Search “Speech”

Meta launches voice and expressive avatar features for the Muse app

On November 6, 2026, Meta announced the launch of voice and expressive avatar features in the Muse app. Users can tap the phone icon in the upper right corner to start a real-time conversation and can interrupt during the conversation. The feature first launched in the United States, covering mobile, web, and Mac, offering hundreds of voice options, and the avatar will make facial expressions and gestures along with the conversation. Meta said voice conversations will also come to AI Glasses and Muse Charm, with the related technologies being Muse Realtime Avatar and Muse Realtime Voice. On the same day, Unite.AI reported that Meta AI Research described the avatar as an audio-driven Diffusion Transformer that outputs 448x768, 25fps vertical video, with about 870 milliseconds from when the user finishes speaking to the first byte of response, and a single GB200 supporting 12 concurrent real-time video generation streams.

Muse launches voice feature Voice

On November 6, 2026, Meta announced the launch of voice and expressive avatar features in the Muse app. Users can tap the phone icon in the upper right corner to start a real-time conversation and can interrupt during the conversation. The feature first launched in the United States, covering mobile, web, and Mac, offering hundreds of voice options, and the avatar will make facial expressions and gestures along with the conversation. Meta said voice conversations will also come to AI Glasses and Muse Charm, with the related technologies being Muse Realtime Avatar and Muse Realtime Voice. On the same day, Unite.AI reported that Meta AI Research described the avatar as an audio-driven Diffusion Transformer that outputs 448x768, 25fps vertical video, with about 870 milliseconds from when the user finishes speaking to the first byte of response, and a single GB200 supporting 12 concurrent real-time video generation streams.

View story:Meta launches voice and expressive avatar features for the Muse app

Hugging Face releases Phonon, an on-device speech model

On November 5, 2026, Hugging Face released Phonon, an on-device TTS model, offering two sizes of approximately 65 MB and 75 MB, covering English, French, German, Spanish, and Portuguese, with more than 100 built-in voices. Phonon 31m supports only English, is open-sourced under Apache 2.0, and can be used in commercial products; Phonon 40m covers five languages and uses a non-commercial license. The company says Phonon ranks first in German and Spanish intelligibility on the Open TTS Leaderboard and leads in CPU streaming first-packet latency benchmarks, and that its accompanying inference framework ptts supports more than 10 backends including CPU, GPU, and NPU. On the same day, Unite.AI reported that Gradium issued a technical follow-up statement saying Phonon surpassed 200,000 downloads within two weeks of launch, and provided 95th-percentile CPU first-audio latency figures: Phonon 31m English at 28 milliseconds, Phonon 40m English at 38 milliseconds, and Pocket TTS at 165 milliseconds; the report also said the open weights were released on October 20.

Hugging Face releases on-device speech model Phonon

On November 5, 2026, Hugging Face released Phonon, an on-device TTS model, offering two sizes of approximately 65 MB and 75 MB, covering English, French, German, Spanish, and Portuguese, with more than 100 built-in voices. Phonon 31m supports only English, is open-sourced under Apache 2.0, and can be used in commercial products; Phonon 40m covers five languages and uses a non-commercial license. The company says Phonon ranks first in German and Spanish intelligibility on the Open TTS Leaderboard and leads in CPU streaming first-packet latency benchmarks, and that its accompanying inference framework ptts supports more than 10 backends including CPU, GPU, and NPU. On the same day, Unite.AI reported that Gradium issued a technical follow-up statement saying Phonon surpassed 200,000 downloads within two weeks of launch, and provided 95th-percentile CPU first-audio latency figures: Phonon 31m English at 28 milliseconds, Phonon 40m English at 38 milliseconds, and Pocket TTS at 165 milliseconds; the report also said the open weights were released on October 20.

View story:Hugging Face releases Phonon, an on-device speech model

Clerk AI adopts LiveKit to power enterprise-grade voice agents

On November 4, 2026, the LiveKit product and engineering blog published an article introducing that Clerk AI has adopted the LiveKit open-source framework and LiveKit Cloud to power enterprise-grade voice agents. The article states that with LiveKit's open-source framework and global edge network, Clerk AI scales its enterprise voice agents to 1,000+ concurrent calls without pre-provisioning, with response times under 500ms. Its ScreenSense achieves 97% detection accuracy for call screening on Apple and Android devices, and TrueReach achieves over 98% accuracy in voicemail detection. The platform natively supports 30+ languages with automatic switching during conversations, has obtained SOC 2 Type II certification, and has built-in TCPA compliance capabilities. There are currently no subsequent developments.

Clerk AI adopts LiveKit to power enterprise-grade voice agents

On November 4, 2026, the LiveKit product and engineering blog published an article introducing that Clerk AI has adopted the LiveKit open-source framework and LiveKit Cloud to power enterprise-grade voice agents. The article states that with LiveKit's open-source framework and global edge network, Clerk AI scales its enterprise voice agents to 1,000+ concurrent calls without pre-provisioning, with response times under 500ms. Its ScreenSense achieves 97% detection accuracy for call screening on Apple and Android devices, and TrueReach achieves over 98% accuracy in voicemail detection. The platform natively supports 30+ languages with automatic switching during conversations, has obtained SOC 2 Type II certification, and has built-in TCPA compliance capabilities. There are currently no subsequent developments.

View story:Clerk AI adopts LiveKit to power enterprise-grade voice agents

Edge0 releases 0.6B voice cloning TTS model Audio8-TTS-Realistic

On October 30, 2026, Edge0 released the open-source-weight voice cloning TTS model Audio8-TTS-Realistic. According to a MarkTechPost report, the model has 601M parameters, supports 23 languages, and its weights and code are released under Apache 2.0. The model requires only one reference audio clip and its corresponding text to perform zero-shot voice cloning, and can also generate speech directly when no reference voice is available. Self-reported results show that its average error rate on the CV3 nine-language benchmark is 4.86%, lower than Qwen3-TTS-1.7B's 7.73%; however, on the MiniMax benchmark it lags behind 4B-class models, with Thai WER reaching 48.57%. There are currently no further developments.

Edge0 releases 0.6B voice cloning TTS model Audio8-TTS-Realistic

On October 30, 2026, Edge0 released the open-source-weight voice cloning TTS model Audio8-TTS-Realistic. According to a MarkTechPost report, the model has 601M parameters, supports 23 languages, and its weights and code are released under Apache 2.0. The model requires only one reference audio clip and its corresponding text to perform zero-shot voice cloning, and can also generate speech directly when no reference voice is available. Self-reported results show that its average error rate on the CV3 nine-language benchmark is 4.86%, lower than Qwen3-TTS-1.7B's 7.73%; however, on the MiniMax benchmark it lags behind 4B-class models, with Thai WER reaching 48.57%. There are currently no further developments.

View story:Edge0 releases 0.6B voice cloning TTS model Audio8-TTS-Realistic

AssemblyAI launches its first text-to-speech model, AssemblyAI-TTS-1

On October 29, 2026, AssemblyAI launched its first text-to-speech model, AssemblyAI-TTS-1, and made it generally available, targeting real-time telephony voice agents, priced at $15 per 1 million characters, with no annual contract and no concurrency tiers. According to a MarkTechPost report, the model is offered only as a hosted API in us-east-1 and eu-north-1, supports HTTP, WebSocket and OpenAI-compatible interfaces, and includes more than 4,000 voices, instant voice cloning and 6 languages; vendor self-testing shows 93.7% entity accuracy and a median time to first byte of 47 milliseconds, but Cekura's independent benchmark measured a streaming latency of 159 milliseconds and a reading accuracy of 81.5%, and at launch it did not yet support SSML, speaking-rate control or emotion tags. On the same day, Artificial Analysis' Provider Voice Arena evaluation showed AssemblyAI-TTS-1 ranked 12th with an Elo of 1227, ranked 16th with a pronunciation robustness score of 81.7%, and generated at 162 characters per second. Later that day, OpenRouter announced that AssemblyAI-TTS-1 was live, targeting voice agent scenarios, with service available from US or EU nodes.