# speech Published articles for speech. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Android Weekly Issue #744 DevFeed: [Android Weekly Issue #744]() Original publisher: [Read original article]() Published: 2026-09-13T16:28:25Z Content type: article Language: en Sources: [Android Weekly]() Topics: [Android](), [Development](), [Compose](), [Artificial Intelligence](), [Rive animation]() Tags: [android](), [android-development](), [compose](), [developers](), [lottie](), [net-11-preview-7](), [offline-speech-recognition](), [rive](), [speech](), [tutorials]() ### AI overview Android Weekly Issue #744 is a developer digest covering Android architecture and AI, annotation-based DSL APIs, animated paywalls, offline speech recognition in keyboards, Compose dialog blurring, and structural hot reload across Android, iOS, and desktop. ### Source excerpt Articles & Tutorials Sponsored Pushwoosh launches a new Push-only plan Stop overpaying for push notifications. Get powerful push infrastructure for $7 per 1,000 MAU, with high-speed delivery, 99.9% push-to-gateway delivery, migration support, and a 24-month price lock. Calculate your savings. Android's AI Era: Hype vs. Architecture Thomas Künneth examines Android's AppFunctions architecture and agent distribution claims behind Google's "intelligent system" framing. Two Racing Lines: Annotations and the DSL in NetFlow 0.7.0 KMP Bits explains adding an annotation-based API layer atop NetFlow's DSL, keeping compile-time-checked mapping and paging support intact. Sponsored How many app issues fell off your radar this week? Mobile teams deal with a firehose of incoming issues: regressions from QA, beta feedback, Play Store reviews, crashes, etc. AI code makes managing this even harder. Triage by Runway pulls every issue into one inbox, de-dupes, assigns, and tracks fixes so nothing is missed. See how it works. Adding Rive and Lottie Animations to Android Paywalls Jaewoong Eum explains building custom Rive and Lottie animated components for RevenueCat's server-driven Android paywalls. I Put Speech Recognition Inside an Android Keyboard. Here's What Broke. Farhad Ranjbar explores the architecture challenges of building an offline speech-recognition engine inside an Android keyboard. How to blur what's behind a Compose dialog, down to API 23 Timofey Krestyanov explains blurring content behind a Compose dialog across window boundaries, down to API 23. Compose HotSwan 2.0.0: The Future of Hot Reload on Android, iOS, and Desktop Jaewoong Eum introduces Compose HotSwan 2.0's own interpreter engine, enabling structural hot reload across Android, iOS, and Desktop. Place a sponsored post Advertise to more than 80k Android developers! Advertise your Android development related service or product! We reach out to more than 80k Android developers around the world, every week, through ## Personalize your product's text-to-speech voice for any language: Fine-tuning with Kubeflow Trainer on Red Hat OpenShift AI DevFeed: [Personalize your product's text-to-speech voice for any language: Fine-tuning with Kubeflow Trainer on Red Hat OpenShift AI]() Original publisher: [Read original article]() Author: Dmytro Hryshchenko, Abhijeet Dhumal Published: 2026-09-09T03:32:28Z Content type: tutorial Language: en Sources: [Red Hat](), [Red Hat Developer]() Topics: [Fine-tuning](), [voice ai](), [Training AI Models](), [Artificial Intelligence](), [distributed-training](), [GPU](), [lora](), [data](), [Open Source]() Tags: [ai](), [artificial-intelligence](), [distributed-training](), [fine-tuning](), [gpu](), [lora](), [open-source](), [speech](), [text-to-speech](), [training](), [voice](), [workflow]() ### AI overview This tutorial explains how to fine-tune the open source Orpheus-3B text-to-speech model for Turkish using Red Hat OpenShift AI and Kubeflow Trainer. It describes packaging distributed training in a TrainJob, scaling across nodes and GPUs, and using LoRA to keep memory usage below 16 GB. The reported result reduces speech errors by more than 90% compared with the base model. ### Source excerpt Can't Read, Won't Buy. That is the title CSA Research gave its survey of 8,709 consumers across 29 countries, and the numbers justify it: 76% prefer to buy in their own language, and 40% will never buy in another. The same rule governs what your product says out loud. The post Personalize your product's text-to-speech voice for any language: Fine-tuning with Kubeflow Trainer on Red Hat OpenShift AI appeared first on Red Hat Developer. ## 5 ways to use Gemini text-to-speech (TTS) in your apps with Firebase AI Logic DevFeed: [5 ways to use Gemini text-to-speech (TTS) in your apps with Firebase AI Logic]() Original publisher: [Read original article]() Author: Ankita Saxena Published: 2026-09-09T00:00:00Z Content type: tutorial Language: en Sources: [Firebase Blog]() Topics: [Firebase](), [voice ai](), [Artificial Intelligence](), [digital accessibility](), [Mobile](), [Web]() Tags: [accessibility](), [ai](), [ai-logic](), [firebase](), [firebase-ai-logic](), [gemini](), [mobile](), [speech](), [text-to-speech](), [web]() ### AI overview The article presents five use cases for Gemini text-to-speech through Firebase AI Logic, including language-learning conversation practice, hands-free content reading, and accessibility-focused audio content. It also describes examples involving Finnish language exercises and recipe instructions. ### Source excerpt News, tutorials, and updates from the Firebase team. ## Integrate Deepgram Flux with Twilio's Conversation Relay DevFeed: [Integrate Deepgram Flux with Twilio's Conversation Relay]() Original publisher: [Read original article]() Author: Dhruv Patel Published: 2026-09-08T00:00:00Z Content type: tutorial Language: en Sources: [Twilio Blog]() Topics: [flux](), [asr](), [Tutorial](), [Latency](), [Node.js](), [WebSocket](), [Streaming]() Tags: [developer-insights](), [flux](), [latency](), [node](), [speech](), [streaming](), [transcription](), [tutorial](), [websocket]() ### AI overview This tutorial shows how to integrate Deepgram Flux with Twilio's Conversation Relay in Node.js. Flux combines speech transcription and turn detection, with configurable end-of-turn confidence, and is described as reducing response latency and false interruptions. ### Source excerpt Integrate Deepgram Flux with Twilio ConversationRelay in Node.js for faster turn detectio, and tunable end-of-turn control. ## Home Assistant 2026.9 adds Modbus device integrations, Cloud voice processing, and Matter network mapping DevFeed: [Home Assistant 2026.9 adds Modbus device integrations, Cloud voice processing, and Matter network mapping]() Original publisher: [Read original article]() Author: Franck Nijhof Published: 2026-09-02T00:00:00Z Content type: release Language: en Sources: [Home Assistant]() Topics: [Home Assistant](), [Release notes](), [Modbus](), [Cloud](), [Matter](), [asr]() Tags: [cloud](), [core](), [home-assistant](), [matter](), [modbus](), [release](), [release-notes](), [security](), [speech](), [ui]() ### AI overview Home Assistant 2026.9 introduces Modbus integrations that can share a connection, new Cloud voice processing focused on speech-to-text, Security dashboard alerts and favorites, shared sources for History and Activity, a Matter network map, keyboard and sound navigation for charts, and 13 new integrations. ### Source excerpt Home Assistant 2026.9! 🎉 I'll be honest: Modbus has never been the flashiest part of Home Assistant. It quietly powers solar inverters, heat pumps, and energy meters, but only if you're comfortable hand-writing your own register map in YAML, a real wall that's kept a huge category of devices out of reach. This release finally makes room on the bus: integrations that already know how a device talks over Modbus, sharing a single connection instead of fighting over it, so you simply pick your device in the UI like anything else in Home Assistant. It won't show up in a screenshot, but modernizing Modbus is hands-down my favorite change this release. 🔧 I'm also really looking forward to jumping on the new Cloud voice processing myself and putting it through its paces. We're a household with a non-English main language, and if you've ever run a private voice assistant in one, you know exactly how fast accents and background noise trip things up. This new speech-to-text engine specifically targets those pain points, so I'm very curious to see how it holds up against my own accent. 😉 And there's plenty more: active alerts and favorites for the Security dashboard, a shared Sources panel for History and Activity, Activity that finally explains why something changed, a brand new network map for Matter, charts you can now navigate by keyboard and sound, and 13 new integrations! 🚀 One more thing before you dive in: we're at IFA Berlin this weekend! It's one of the largest consumer electronics expos out there, running September 4 to 8, 2026, and the Open Home Foundation has its own booth in the Smart Home hall: Hall 1.2, stand 1.2-153. If you're around, come find us and say hi. 👋 Enjoy the release! ../Frenck Home Assistant Cloud Seeing through the Cloud Test the new voice processing Float in to be a Cloud subscriber Active alerts and favorites for the Security dashboard A new Sources pane for History and Activity From what changed to why it changed Insights into your Matter world ## The Open ASR Leaderboard Adds Its First Global South Language DevFeed: [The Open ASR Leaderboard Adds Its First Global South Language]() Original publisher: [Read original article]() Author: Eric Bezzam; Shobhit Banga; Manas Dhir; Bhaskar Singh; Manmeet Kaur; Aaditya Pareek; Walecha; Sagar Jain; Hanuman Sidh; Vanshika Chhabra Published: 2026-08-28T00:00:00Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [asr](), [Benchmark](), [Human-AI evaluation]() Tags: [artificial-intelligence](), [asr](), [audio](), [benchmark](), [benchmarks](), [contributors](), [datasets](), [devices](), [evaluation](), [leaderboard](), [metrics](), [open](), [open-source](), [research](), [speech]() ### AI overview The Open ASR Leaderboard introduces Monsoon evaluation sets for Hindi in India, expanding coverage beyond European languages and testing how recognition performance varies across populations and conditions. The sets use public and private splits, speaker-disjoint data, detailed speaker attributes, and variation in geography, age, gender, vocabulary, devices, acoustic environments, speech type, speech rate, and transcript validity. ### Source excerpt We're on a journey to advance and democratize artificial intelligence through open source and open science. ## Intelligent transcription with Gemini 3.5 Transcribe DevFeed: [Intelligent transcription with Gemini 3.5 Transcribe]() Original publisher: [Read original article]() Author: Diego Melendo Casado Published: 2026-08-26T17:01:00Z Content type: release Language: en Sources: [Google DeepMind News]() Topics: [asr](), [voice ai](), [Google AI](), [API](), [real-time](), [Streaming](), [Latency](), [Artificial Intelligence]() Tags: [ai](), [analysis](), [analytics](), [android](), [api](), [audio](), [developer](), [developers](), [gemini](), [google](), [google-ai](), [latency](), [macos](), [model](), [none](), [real-time](), [speech](), [streaming](), [transcription](), [voice](), [workflows]() ### AI overview Google introduces Gemini 3.5 Transcribe, a speech-to-text model for accurate, formatted transcription in noisy and specialized audio. Developers can use it through the Gemini API in Google AI Studio and Gemini Enterprise Agent Platform for voice agents, real-time captioning, and post-call analytics. It supports real-time streaming and pre-recorded audio processing, custom vocabulary, speaker attribution, word-level timestamps, and more than 85 languages. ### Source excerpt Now you can get more intelligent speech-to-text transcription with Gemini 3.5 Transcribe. ## Gemini 3.5 Transcribe now available on AI Gateway DevFeed: [Gemini 3.5 Transcribe now available on AI Gateway]() Original publisher: [Read original article]() Author: Jerilyn Zheng Published: 2026-08-26T00:00:00Z Content type: release Language: en Sources: [Vercel News]() Topics: [Artificial Intelligence](), [Google](), [Streaming](), [SDKs](), [vercel ai sdk](), [WebSocket]() Tags: [ai](), [ai-gateway](), [audio](), [gemini](), [google](), [sdk](), [speech](), [streaming](), [transcription]() ### AI overview Vercel announces that Google's Gemini 3.5 Transcribe models are available through AI Gateway for both complete recordings and live audio. The models support multilingual detection across more than 85 languages, custom vocabulary, speaker identification, word-level timestamps, and streaming transcription through AI SDK 7. ### Source excerpt Gemini 3.5 Transcribe from Google is now available on AI Gateway for recorded and live audio: google/gemini-3.5-transcribe transcribes a complete audio file in one request. google/gemini-3.5-transcribe-live transcribes audio over a WebSocket and returns text as the audio arrives. Both models automatically detect more than 85 languages, including when a speaker switches languages. You can also provide custom vocabulary to improve the transcription of names, technical terms, and uncommon spellings. The model for complete recordings can also identify speakers and return word-level timestamps. Streaming transcription is available in AI SDK 7. Install the latest AI SDK and AI Gateway provider: Transcribe live audio Use streamTranscribe with a ReadableStream of raw audio chunks. Set inputAudioFormat to match the audio being sent: Transcribe a complete recording Use transcribe to send a complete audio file and receive the finished transcript: Try Gemini 3.5 Transcribe Live in the model playground, browse all transcription models, or read the speech quickstart. Read more ## Measuring benchmark optimization in speech recognition DevFeed: [Measuring benchmark optimization in speech recognition]() Original publisher: [Read original article]() Author: Theo Lebryk; Eric Bezzam; Alice; David Ayllon; Jakub Piotr Cłapa; Jens Madsen; Panagiotis Tzirakis Published: 2026-08-21T00:00:00Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [asr](), [Benchmark](), [benchmark overfitting machine learning](), [Machine learning](), [datasets](), [Open Source]() Tags: [asr](), [audio](), [benchmark](), [datasets](), [errors](), [measurement](), [model](), [open-source](), [research](), [speech](), [transcription]() ### AI overview The article examines benchmark optimization, or "benchmaxxing," in speech recognition. It presents three tests and evaluates 11 open-source ASR models, finding that some reproduced benchmark transcripts even when the audio contradicted them. The research also uses model ensembles and human annotations to identify and validate likely benchmark errors. ### Source excerpt We're on a journey to advance and democratize artificial intelligence through open source and open science. ## Fish Audio models now available on Vercel AI Gateway for free DevFeed: [Fish Audio models now available on Vercel AI Gateway for free]() Original publisher: [Read original article]() Author: Jerilyn Zheng Published: 2026-08-19T00:00:00Z Content type: release Language: en Sources: [Vercel News]() Topics: [voice ai](), [Vercel](), [asr](), [Streaming](), [Low Latency](), [SDKs]() Tags: [ai-gateway](), [audio](), [browser](), [free](), [launch](), [models](), [sdk](), [speech](), [streaming](), [text-to-speech](), [transcription](), [voice]() ### AI overview Fish Audio's text-to-speech and transcription models are available on Vercel AI Gateway. The models are free for 30 days, with AI SDK 7 support for speech generation and transcription, including timestamped segments and word-level timing. ### Source excerpt Fish Audio's audio models are now available on AI Gateway. To celebrate the launch, every Fish Audio model is free on AI Gateway for the next 30 days, through September 18. Capability Regular Through September 18 Text-to-speech $15.00 per million characters Free Speech-to-text $0.36 per hour of audio Free Four models from Fish Audio are available, including their latest text-to-speech model: fish-audio/s2.1-pro (text-to-speech): Built for low-latency streaming; clones a voice from a reference recording. fish-audio/transcribe-1 (transcription): Returns the text along with the duration of the audio and timestamped segments, down to individual words. fish-audio/s2-pro (text-to-speech): Covers around eighty languages and takes inline tags, plain-language directions written into the text itself, so you can change how a single word or phrase is delivered instead of setting one style for the whole request. fish-audio/s1 (text-to-speech): Reads text that can carry markers for emotion, tone, and sound effects. How to use models during the offer period Using the standard model name (i.e., fish-audio/s2.1-pro) is free, but will automatically begin billing when the offer period ends. To ensure you aren't billed after the free period, add the -free suffix to the standard name, and the model will stop serving when the offer ends (i.e., fish-audio/s2.1-pro-free). Speech and transcription ship in the current AI SDK 7 release. Text-to-speech Generate spoken audio from text with generateSpeech and write the result: Speech-to-text Transcribe recordings into text with transcribe. The audio can be a buffer, a base64 string, or a URL: Each segment carries the text and its start and end time in seconds, down to individual words. Playground You can also try the Fish Audio models without writing any code. Open the models list, click into a model, and send text or audio to hear or read the result in your browser. For a full overview of how to utilize audio models, refer to the speech quickst ## Build Low-Latency Multilingual Voice Agents: Open Weights & Full Deployment Control with NVIDIA Magpie TTS DevFeed: [Build Low-Latency Multilingual Voice Agents: Open Weights & Full Deployment Control with NVIDIA Magpie TTS]() Original publisher: [Read original article]() Author: Maryam Motamedi; Mikyas Desta; Jason Li; Jason Roche Published: 2026-08-10T16:25:36Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [Low Latency](), [Latency](), [voice ai](), [NVIDIA NIM](), [Open Source Models & Datasets](), [asr](), [Conversational AI](), [Large Language Model](), [Automation]() Tags: [agents](), [ai](), [api](), [audio](), [automation](), [conversational-ai](), [data](), [developers](), [infrastructure](), [latency](), [llm](), [low-latency](), [model](), [models](), [nvidia](), [nvidia-nim](), [open](), [speech](), [text-to-speech](), [voice]() ### AI overview This developer article presents NVIDIA Magpie Multilingual TTS as an open-weights, 364M-parameter text-to-speech model for building low-latency multilingual voice applications. It explains how a self-managed cascaded ASR, TTS, and LLM architecture can provide deployment control, tuning, privacy, and predictable performance across 12 languages. ### Source excerpt Build Low-Latency Multilingual Voice Agents: Open Weights & Full Deployment Control with NVIDIA Magpie TTS Every voice interaction has a latency budget. By the time a user hears your application respond, you've already spent precious milliseconds capturing audio, transcribing speech, running an LLM, retrieving context, and generating a response. Text-to-speech (TTS) is the final step -- and the one users notice most. If speech generation is slow, the whole experience feels slow. ## How we built a realtime system for responsive voice AI in six months DevFeed: [How we built a realtime system for responsive voice AI in six months]() Original publisher: [Read original article]() Published: 2026-08-03T07:00:00Z Content type: article Language: en Sources: [OpenAI News]() Topics: [voice ai](), [Low Latency](), [Inference](), [model-deployment](), [systems](), [Artificial Intelligence]() Tags: [agents](), [ai](), [architecture](), [audio](), [chatgpt](), [engineering](), [gpt](), [inference](), [latency](), [llm](), [llms](), [low-latency](), [model](), [speech](), [systems](), [text-to-speech](), [voice](), [voice-ai]() ### AI overview OpenAI describes GPT-Live, a full-duplex voice system designed for continuous, natural conversation. Its low-latency architecture streams audio in both directions, separates asynchronous delegation from the core voice path, and coordinates stateful inference, dynamic context management, and protocol-level optimization. ### Source excerpt GPT-Live enables continuous voice interaction with AI, using a turnless speech model and low-latency architecture for faster, more natural conversations. ## How avatarin built a 24/7 retail agent with GPT-Realtime DevFeed: [How avatarin built a 24/7 retail agent with GPT-Realtime]() Original publisher: [Read original article]() Published: 2026-07-30T00:00:00Z Content type: article Language: en Sources: [OpenAI News]() Topics: [AI Chat](), [AI Bots](), [Retrieval Augmented Generation (RAG)]() Tags: [agent](), [ai](), [ai-agent](), [api](), [customer](), [gpt](), [low-latency](), [openai](), [retail](), [retrieval](), [retrieval-augmented-generation](), [speech](), [voice]() ### AI overview avatarin built a 24/7 multilingual retail shopping agent for Yamada Denki using OpenAI's GPT-Realtime. The agent supports voice conversations and product recommendations; a two-week public campaign reported about 30,000 users and 92% positive survey responses. ### Source excerpt avatarin uses OpenAI's GPT-Realtime to give Yamada Denki shoppers 24/7 multilingual support. In two weeks, 30,000 people used the agent and 92% of survey responses were positive. ## Grok Voice Think Fast 2.0 now available on AI Gateway DevFeed: [Grok Voice Think Fast 2.0 now available on AI Gateway]() Original publisher: [Read original article]() Author: Jerilyn Zheng Published: 2026-07-29T00:00:00Z Content type: release Language: en Sources: [Vercel News]() Topics: [speech-to-speech](), [Artificial Intelligence](), [SDKs](), [API](), [API keys]() Tags: [ai-gateway](), [audio](), [latency](), [model](), [playground](), [reasoning](), [sdk](), [server](), [speech](), [speech-to-speech](), [tokens](), [tool](), [transcription](), [voice]() ### AI overview Grok Voice Think Fast 2.0 from xAI is now available through AI Gateway as a speech-to-speech model. It processes audio input and output while reasoning in parallel, using fewer reasoning tokens to reduce tool-call delays. The article also describes transcription performance in noisy and compressed audio conditions and access through the AI SDK realtime API. ### Source excerpt Grok Voice Think Fast 2.0 from xAI is now available on AI Gateway. It is a speech-to-speech voice model that takes audio in and audio out, improving on the previous Grok Voice model in reasoning, transcription accuracy, and conversation. The model reasons in parallel with speech, so it can think through a query while talking without adding latency. It has also been trained to use fewer reasoning tokens than before, so tool calls fire sooner, often before the end of the agent's first sentence. Transcription holds up in real-world conditions, including background noise and telephony compression. Use xai/grok-voice-think-fast-2.0 through the AI SDK's realtime API. Mint a short-lived token on the server so your API key never reaches the client: See the realtime documentation to build a voice agent with Grok Voice Think Fast 2.0. Try out the model in the AI Gateway playground. Read more ## AI Gateway now supports streaming transcription DevFeed: [AI Gateway now supports streaming transcription]() Original publisher: [Read original article]() Author: Jerilyn Zheng Published: 2026-07-22T00:00:00Z Content type: release Language: en Sources: [Vercel News]() Topics: [Streaming](), [asr](), [SDKs](), [vercel ai sdk](), [Latency](), [Whisper](), [Artificial Intelligence]() Tags: [agent](), [agents](), [gpt](), [latency](), [openai](), [sdk](), [speech](), [streaming](), [streams](), [transcription](), [transcripts](), [voice](), [whisper]() ### AI overview AI Gateway now supports beta streaming transcription through the AI SDK's streamTranscribe function. Applications can stream audio as it is captured and receive partial and final transcript updates with low latency, enabling use cases such as live captioning, voice input, and voice-enabled agents. ### Source excerpt AI Gateway now supports streaming transcription. Previously, transcription required a complete audio file and returned the full transcript in a single response. Now you can stream audio in as it's captured and get transcript updates back as the model produces them, keeping latency low for uses like live captioning and voice input. Streaming transcription is in beta and available through the AI SDK's streamTranscribe function with any streaming-capable transcription model. The example below streams raw PCM audio to openai/gpt-realtime-whisper and prints each transcript delta as it arrives. The result stream also carries partial and final transcripts: The same code works across providers: swap the model string to use xai/grok-stt or any other streaming-capable transcription model. Streaming transcription also makes it easy to add a voice mode to an agent. Stream the user's speech to a transcription model and pass the live text to your agent. The agent itself does not change: it still receives text, so this works with any text-based agent. For agents that speak back, pair it with speech generation, or use realtime voice for full two-way conversation. For more information on streaming transcription on AI Gateway, see the documentation. Read more ## Building AI Features in Kotlin Multiplatform Apps with Local Prompt Iteration DevFeed: [Building AI Features in Kotlin Multiplatform Apps with Local Prompt Iteration]() Original publisher: [Read original article]() Author: Paul Woitaschek Published: 2026-07-17T11:49:08Z Content type: tutorial Language: en Sources: [Stories by Paul Woitaschek on Medium]() Topics: [Kotlin Multiplatform](), [LLMs](), [Mobile](), [API](), [Back end](), [voice ai](), [anthropic](), [OpenAI]() Tags: [ai](), [anthropic](), [api](), [backend](), [kotlin](), [kotlin-multiplatform](), [llm](), [llms](), [mobile](), [mobile-app-development](), [openai](), [speech](), [voice]() ### AI overview This tutorial presents a workflow for building AI features in Kotlin Multiplatform mobile apps. It recommends defining the app interface first, implementing the feature locally with Koog to iterate quickly on prompts, and moving the implementation to a backend once the output is reliable. The example uses natural-language and voice-based nutrition tracking, with LLM output converted into proposed structured actions. ### Source excerpt TLDR: Define the interface in the app, iterate locally with Koog, move the logic to the backend when it's done. AI features have a weird reputation among mobile developers. The model sits behind an API, the prompt lives on a server someone else owns, and every change means a ticket, a review, a deploy. So the whole topic feels like backend territory. But an AI feature is just a feature. Text in, structured data out, some logic in between. The only unusual part: the core of that logic is a prompt. And prompts are not designed, they are iterated. You try an input, look at the output, adjust, try again. Dozens of times. So the real problem is not building the feature. It's making that loop fast. The pattern I would like to show: define an interface in the app, build the real UI against it, implement it locally with Koog, iterate until the output is solid, then move the implementation to the backend. In Kotlin that last step is mostly copy paste. The example: track everything in one sentence We build Yazio, a nutrition tracking app, as Kotlin Multiplatform: one shared codebase for iOS and Android. That matters here, because everything below is written once and runs on both platforms. Tracking today is many small interactions: search a food, pick a portion, add water, log a workout. Now imagine the user just says what happened: "I had two scrambled eggs and a coffee for breakfast, drank half a liter of water, went for a 30 minute run, and for lunch I ate the same as yesterday."Users can track using voice The app answers with a list of proposed actions. The user deletes the wrong ones, speaks again to add what's missing, and confirms everything at once. Speech to text is solved on both platforms, so from our side the input is just a string. Messy natural language in, exact typed data out. That's what LLMs are good at. It's also the kind of logic you won't get right on the first try. The constraint: the key can't live in the app You can't ship an OpenAI or Anthropic API ke ## ChatGPT Live and the New Architecture of Voice AI DevFeed: [ChatGPT Live and the New Architecture of Voice AI]() Original publisher: [Read original article]() Author: RisingStack Engineering Published: 2026-07-16T12:38:22Z Content type: article Language: en Sources: [RisingStack]() Topics: [voice ai](), [Artificial Intelligence](), [ChatGPT](), [OpenAI](), [systems]() Tags: [ai](), [architecture](), [audio](), [chatgpt](), [engineering](), [gpt](), [here](), [interrupt](), [messages](), [models](), [openai](), [quality](), [speaking](), [speech](), [systems](), [talk](), [tool](), [update](), [voice-ai](), [voices]() ### AI overview The article explains how OpenAI's GPT-Live models power ChatGPT Voice through a full-duplex architecture. The system can listen and speak simultaneously, handle interruptions, pause while users think, and delegate deeper reasoning, web search, or other complex work to another model. ### Source excerpt OpenAI has introduced GPT-Live, a new generation of voice models that now powers ChatGPT Voice. At first, this may sound like another voice-quality update. The voices have been remastered, ChatGPT should interrupt less often, and it can respond more naturally when you pause, change direction or speak over it. But GPT-Live is more than a [...] The post ChatGPT Live and the New Architecture of Voice AI appeared first on RisingStack Engineering. ## Introducing Real World VoiceEQ: Measuring the human quality of voice AI DevFeed: [Introducing Real World VoiceEQ: Measuring the human quality of voice AI]() Original publisher: [Read original article]() Author: David Ayllon; Alice; Jeff Brooks; Franc Camps Febrer; Jakub Piotr Cłapa; Theo Lebryk; Jens Madsen; Olya Ossipova; Sharath Rao; Hoon Shin Published: 2026-07-15T00:00:00Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [voice ai](), [Benchmark](), [Human-AI evaluation](), [Artificial Intelligence](), [asr](), [human feedback](), [Open Source]() Tags: [ai](), [announcement](), [audio](), [benchmark](), [benchmarks](), [evaluation](), [human-feedback](), [leaderboard](), [open-source](), [reinforcement-learning](), [speech](), [voice-ai]() ### AI overview Real World VoiceEQ is a benchmark for evaluating the human quality of voice AI beyond latency and word error rate. It measures how voice systems recognize, produce, and respond to acoustic information such as tone, emotion, speaker identity, and background context across ASR, TTS, speech-to-speech, and speech understanding. The benchmark covers more than 40 voice models, 15+ evaluation dimensions, and more than 60 metrics, using over 1 million human ratings. ### Source excerpt We're on a journey to advance and democratize artificial intelligence through open source and open science. ## Build realtime voice agents on AI Gateway DevFeed: [Build realtime voice agents on AI Gateway]() Original publisher: [Read original article]() Author: Kevin Dawkins Published: 2026-06-29T07:00:00Z Content type: tutorial Language: en Sources: [Vercel News]() Topics: [vercel ai sdk](), [Language models](), [SDKs]() Tags: [agents](), [ai-gateway](), [audio](), [observability](), [openai](), [routing](), [speech](), [streaming](), [text-to-speech](), [tools](), [transcription](), [voice]() ### AI overview Vercel AI Gateway adds beta support for realtime voice, text-to-speech, and speech-to-text in AI SDK 7. The article explains browser-based realtime sessions, token-based authentication, interruption handling, tool calls, and direct audio input/output models. ### Source excerpt AI Gateway now supports audio/voice. You can add realtime voice, text to speech, and speech to text with the same calls you already use for text, image, and video, routed through AI Gateway alongside every other modality. Audio launches with models from OpenAI and xAI. Each call gets the same provider routing, observability, spend controls, and bring-your-own-key support you already use for your other models. These capabilities are in beta and available in AI SDK 7. Capability How it works Use it for Realtime voice Live audio in and out, for streaming, low-latency session Two-way voice agents and live conversation Text to speech Text in, audio file out, single request Voiceovers, spoken responses, audio versions of written content Speech to text Recorded audio in, text out, single request Transcribing voice notes, call recordings Getting started Realtime, speech, and transcription model are supported on AI SDK 7. Realtime voice agents Realtime turns your app into something a user can hold a conversation with. When they speak, the model responds right away. Because it replies in the moment instead of waiting for a full turn, users can interrupt and talk over it the way they would with a person. It fits voice assistants, customer support agents, hands-free tools, and anywhere a user would rather talk than type. What sets it apart from chaining models together is that a single realtime model hears audio and produces audio directly, instead of running a speech-to-text, then language model, then text-to-speech pipeline. In the browser, the useRealtime hook manages the WebSocket connection, microphone capture, and audio playback. The connection is authenticated with your AI Gateway credential, so you mint a short-lived token on the server and hand the browser only that token. Your API key never reaches the client. Add a route that mints the token: Then connect from a client component: The hook captures the microphone, streams the audio to the model through AI Gateway, and ## Realtime voice, speech, and transcription now supported on AI Gateway DevFeed: [Realtime voice, speech, and transcription now supported on AI Gateway]() Original publisher: [Read original article]() Author: Jerilyn Zheng Published: 2026-06-29T00:00:00Z Content type: release Language: en Sources: [Vercel News]() Topics: [real-time](), [voice ai](), [vercel ai sdk](), [Artificial Intelligence](), [SDKs](), [WebSocket](), [browser](), [Low Latency](), [API keys](), [observability](), [App]() Tags: [agent](), [ai](), [api](), [app](), [audio](), [browser](), [code](), [documentation](), [latency](), [low-latency](), [playground](), [real-time](), [sdk](), [speech](), [transcription](), [voice]() ### AI overview AI Gateway adds beta support for realtime voice and audio models through AI SDK 7, enabling voice agents, text-to-speech, and speech-to-text. Developers can use the realtime example, quickstart, or browser playground, with observability, spend controls, bring-your-own-key support, and no markup or platform fees. ### Source excerpt AI Gateway now supports voice and audio models. You can build realtime voice agents, generate speech from text, and transcribe audio to text. This provides the same observability, spend controls, and bring-your-own-key support as text, image, and video models in AI Gateway, with no markup or platform fees. These capabilities are in beta and available via AI SDK 7. With realtime support, a single model takes audio in and audio out, so a user can talk and hear a reply back in near real time instead of waiting on a chain of separate models. Capability What it does Realtime voice agents Model listens to the user, works out a response, and speaks it back in a live, low-latency conversation. It can call your tools mid-conversation to look something up or take an action. The useRealtime hook handles microphone capture and playback. Text to speech Generate spoken audio from text, with a selectable voice and output format such as MP3. Use it for voiceovers, audio versions of written content, and spoken responses. Speech to text Transcribe recordings into text, from a file buffer, base64 string, or URL. Use it for voice notes or other transcriptions. Two ways to get started: Follow the realtime example below or the realtime quickstart to add a voice agent to your app. Use the playground. Talk to a realtime model in the browser, no code required, in the AI Gateway Playground. Realtime example A voice agent has two pieces: a server route that mints a short-lived token, so your API key never reaches the client, and a browser component that connects with it. Add the token route: Then connect from the browser. The useRealtime hook fetches that route and manages the WebSocket connection, microphone capture, and audio playback: Playground You can also try audio models without writing any code. Open the models page, click into a model, and interact with it right in the browser: Talk to a realtime model to hold a voice conversation Send text and have a transcription model read it back ## xAI Grok audio models now available on Vercel AI Gateway DevFeed: [xAI Grok audio models now available on Vercel AI Gateway]() Original publisher: [Read original article]() Author: Carlton Aikins Published: 2026-06-29T00:00:00Z Content type: release Language: en Sources: [Vercel News]() Topics: [vercel ai sdk](), [voice ai](), [Artificial Intelligence](), [SDKs](), [API](), [WebSocket](), [Vercel](), [React](), [browser]() Tags: [ai](), [ai-gateway](), [api](), [audio](), [browser](), [models](), [observability](), [react](), [release](), [responses](), [routing](), [sdk](), [server](), [speech](), [text-to-speech](), [vercel](), [voice]() ### AI overview xAI Grok audio models are now available through Vercel AI Gateway and the AI SDK 7 release. The integration supports realtime voice, text-to-speech, and speech-to-text, with routing, observability, and spend controls. ### Source excerpt xAI's audio models are now live on AI Gateway. Realtime voice, text to speech, and speech to text are all available through the AI SDK with the same routing, observability, and spend controls as your other models. These capabilities are available on the AI SDK 7 release. Available models Capability Models Realtime voice xai/grok-voice-think-fast-1.0 Text to speech xai/grok-tts Speech to text xai/grok-stt Realtime A voice agent has two pieces: a server route that mints a short-lived token, so your API key never reaches the client, and a browser component that connects with it. Add the token route: this example sets model to xai/grok-voice-think-fast-1.0: Then connect from the browser. The useRealtimehook from @ai-sdk/react fetches that route and manages the WebSocket connection, microphone capture, and audio playback: Text to speech Generate spoken audio from text with generateSpeech. Pass a voice and an output format, then write the result to a file with xai/grok-tts: Speech to text Transcribe recordings into text with transcribe. This example uses xai/grok-stt: Playground You can also try the xAI audio models directly in the AI Gateway playground. Open the models list and click into any of the models to use them directly in the browser. The xai/grok-voice-think-fast-1.0 playground here allows you to talk to the agent and see responses instantly: More information Realtime quickstart Speech quickstart See all xAI models Read more ## Introducing the FFASR Leaderboard: Benchmarking ASR in the Real World DevFeed: [Introducing the FFASR Leaderboard: Benchmarking ASR in the Real World]() Original publisher: [Read original article]() Author: Daniel Gert Nielsen; Shivam Saini; Alessia Milo; Georg Götz; Eric Bezzam Published: 2026-06-24T00:00:00Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [asr](), [benchmarking](), [Benchmark](), [Open Source]() Tags: [announcement](), [asr](), [audio](), [benchmarking](), [community](), [deployment](), [development](), [evaluation](), [leaderboard](), [open](), [open-source](), [partnerships](), [performance](), [speech]() ### AI overview Treble Technologies and Hugging Face introduce the open, community-driven FFASR Leaderboard, a benchmark for evaluating automatic speech recognition models in realistic far-field acoustic conditions. It measures performance across factors such as reverberation, background noise, microphone distance, and low signal-to-noise ratios, while also showing the tradeoff between recognition accuracy and speed. ### Source excerpt We're on a journey to advance and democratize artificial intelligence through open source and open science. ## Fluid, natural voice translation with Gemini 3.5 Live Translate DevFeed: [Fluid, natural voice translation with Gemini 3.5 Live Translate]() Original publisher: [Read original article]() Author: Anuda Weerasinghe Published: 2026-06-09T15:16:25Z Content type: release Language: en Sources: [Google DeepMind News]() Topics: [Google AI](), [Language models]() Tags: [ai](), [api](), [audio](), [developer](), [gemini](), [google](), [google-ai](), [none](), [real-time](), [speech](), [speech-to-speech](), [streaming](), [voice]() ### AI overview Google announces Gemini 3.5 Live Translate, an audio model for continuous speech-to-speech translation that preserves vocal qualities across more than 70 languages. Developers can access it in public preview through the Gemini Live API and Google AI Studio. ### Source excerpt Gemini 3.5 Live Translate brings near real-time, natural speech translation to Google AI Studio, Google Translate and Google Meet. ## Reachy Mini goes fully local DevFeed: [Reachy Mini goes fully local]() Original publisher: [Read original article]() Author: Amir Mahla; Andres Marafioti Published: 2026-05-27T00:00:00Z Content type: tutorial Language: en Sources: [Hugging Face - Blog]() Topics: [reachy](), [gemma4](), [llama.cpp](), [Large Language Model](), [speech-to-speech](), [hugging face](), [WebSocket]() Tags: [api](), [artificial-intelligence](), [audio](), [blog](), [gemma](), [guide](), [how-to](), [llama](), [llama-cpp](), [llm](), [local](), [local-ai](), [open-source](), [reachy](), [robotics](), [server](), [speech](), [speech-to-speech]() ### AI overview A tutorial for running fully local conversations with a Reachy Mini robot. It describes a cascaded VAD, speech-to-text, LLM, and text-to-speech pipeline using llama.cpp with Gemma 4, Silero VAD, Parakeet-TDT STT, and Qwen3-TTS, connected through a Realtime API-compatible WebSocket. The setup avoids cloud services, API keys, and sending data off the local machine. ### Source excerpt We're on a journey to advance and democratize artificial intelligence through open source and open science. [Next page]()