Tag
This paper presents Wayu-Paxa-TTS-Edge, an 82M-parameter Thai TTS model trained on synthetic speech from a voice-cloning teacher, achieving high accuracy and prosody for on-device use without reference audio.
The article discusses hidden sources of latency in voice-agent systems after streaming is enabled and seeks advice on which performance metrics are crucial for real-time agents.
Gradium has released a new Text-to-Speech model that combines high accuracy with low latency, achieving under 250ms time-to-first-audio through intelligent contextual handling and silence trimming.
Introducing the local voice cloning tool VoiceStudio, which only requires 8 seconds of recording to highly replicate the voice, runs completely locally to protect privacy, and supports multiple languages and platforms.
A lightweight TTS implementation, replicating Audio8's training with 2000 hours of data, achieving a SIM metric of 0.72 in under 10 hours of training on H200.
TontaubeV1 is an open-weight text-to-speech model released for local long-form generation, supporting English and German with zero-shot voice cloning and low-latency inference on GPUs.
The Bland team explores the uncanny valley effect of AI voices and develops TTS models that can simulate human imperfections to enhance the naturalness and human-like experience of phone interactions.
audio.cpp version 0.7 introduces a new Arena UI for comparing audio models locally, expands to 62 model families with over 85 variants, and supports deployment on edge hardware like NVIDIA Jetson Orin.
Cartesia's TTS model Sonic-3.6 is now generally available, having been rebuilt from the architecture up and receiving positive feedback from its preview phase.
A new meta-ranking combines three public TTS leaderboards into one unified ranking of 110 models across 46 providers, updated weekly with a fixed methodology.
Breeze TTS 2 has been released on Hugging Face as the leading open weights text-to-speech model, supporting 50 languages and advanced features like voice design and direction.
Kyutai Labs has open-sourced their Pocket TTS training stack, including data pipeline, recipes, and evaluations, allowing developers to train text-to-speech models on GPUs and run them on CPUs.
Audiobookify is an EPUB reader app for iPhone that features local offline TTS with optimized AI models for efficiency in battery life and narration quality.
FireRedAudio and FireRedTTS3 are AI models for unified audio understanding and generation, offering ASR, TTS, voice cloning, editing, and multilingual support with competitive benchmarks.
Tencent has open-sourced AuK, a 1.5B foundation model for speech generation and editing that supports zero-shot TTS, content editing, and more through a unified natural-language instruction interface.
This project converts ebooks to audiobooks, supporting various formats and TTS engines, including voice cloning, ideal for listening during commutes.
Soniox TTS v2 is a new text-to-speech model offering premium voice quality, expressive control via audio tags, high-fidelity voice cloning, support for 60+ languages, and low-latency streaming, priced at $0.70 per generated hour.
A technical breakdown of an enterprise Voice AI architecture that cuts telephony costs by 40-60% via wholesale carriers and achieves sub-500ms latency using Deepgram, Claude/GPT-4o-mini, and ElevenLabs/Cartesia, orchestrated through n8n and Supabase.
Introducing KrillinAI, a free locally-run video translation tool that supports precise subtitles, natural voiceover, and voice cloning. It integrates Whisper, LLM, and CosyVoice, and supports Win/Mac and yt-dlp.
VideoLingo is an open-source video translation, localization, and dubbing tool that uses WhisperX and AI to deliver Netflix-level subtitles and multilingual dubbing. It supports downloading via yt-dlp and multiple TTS options, helping video reposters automate the entire workflow.