AI Briefing#speech-synthesis
The latest AI and developer news about #speech-synthesis, with the original source and a short summary.
The latest AI and developer news about #speech-synthesis, with the original source and a short summary.
The release includes Mini, Micro, and Atom variants, with the 34.31M parameter Mini model supporting reference voice tuning and 48 kHz enhancement on CPU and CUDA.
The 74M parameter codec uses Grouped FSQ to enable parallel prediction, outperforming Mimi and NanoCodec in reconstruction quality and speech generation metrics.
The 131M parameter model supports 6 languages and 14 voices, achieving 30ms warm start on M5 Max and 0.85% WER on the Open TTS Leaderboard.
Researchers at UC Berkeley and UCSF developed a streaming system that produces the first sound of a sentence within about one second of detecting a brain signal, compared to an eight-second delay for a full sentence in their earlier approach.
The upgraded weights rank #1 open-weight and #6 overall in Provider Voice, while also leading in Controlled Voice and Pronunciation Robustness.
The system uses BookLLM to profile characters with 97% accuracy in English and 94% in Spanish, enabling distinct and consistent voices across full-length books.
ElevenLabs supports SSML tags like <break> and <phoneme> on older models, while Eleven v3 and v4 replace them with natural language Audio Tags.
Watermarking provides definitive verification for specific providers like ElevenLabs, while statistical tools offer probabilistic detection across broader sources.
The integration includes MAI-Voice-2.1, MAI-Voice-2.1-Flash, and MAI-Transcribe-2 Streaming with no platform fee or markup on inference.
Tontaube has released TontaubeV1, a 2.9B parameter open-source TTS model specialized for long-form generation and low-latency inference.
Prosody, comprising rhythm, intonation, and stress patterns, is a core element of TTS models that determines the human-likeness of AI voices.
Neural TTS is driving market growth by replacing traditional methods with deep learning-based natural speech synthesis.
Apple has unveiled a new audio synthesis architecture that generates high-quality speech in real time while minimizing memory usage.
A fine-tuning toolkit has been released that lets the ultra-compact TTS model Inflect-Nano-v2 be trained on a specific voice or language.
Inflect-Micro-v2, a highly efficient TTS model capable of local execution with fewer than 10 million parameters, has been released.
Neuphonic has open-sourced NeuTTS-2E, a 125M-parameter on-device TTS model capable of controlling 7 emotions.
The RTCA workshop at NeurIPS 2026, covering research on real-time multimodal conversational agents, has opened its call for papers.
A Continuous Diffusion-based speech model was shown to follow scaling laws while demonstrating potential for efficient inference.
Supertonic 3 has been released as an on-device TTS supporting 31 languages and emotion tags.
Flare-TTS 28M, trained for 24 hours on a single A6000, has been released.
Microsoft has open-sourced its VibeVoice ASR and Realtime models.
Google Cloud has selected ElevenLabs as its 2026 Applied AI Partner.
xAI has launched real-time and batch speech recognition and speech synthesis APIs.
A 2B-scale TTS/voice cloning model supporting 30 languages and 48kHz output.