We’re on a journey to advance and democratize artificial intelligence through open source and open science.

YouTube video by Qwen

"Robot Voices" is a downloadable sound collection of robotic speech. 50 Robots, many unique expressions per robot, delivered as 96 kHz / 24 bit WAV files.

"Robot Voices" is a downloadable sound collection of robotic speech. 50 Robots, many unique expressions per robot, delivered as 96 kHz / 24 bit WAV files.

"Robot Voices" is a downloadable sound collection of robotic speech. 50 Robots, many unique expressions per robot, delivered as 96 kHz / 24 bit WAV files.

"Robot Voices" is a downloadable sound collection of robotic speech. 50 Robots, many unique expressions per robot, delivered as 96 kHz / 24 bit WAV files.

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

"Robot Voices" is a downloadable sound collection of robotic speech. 50 Robots, many unique expressions per robot, delivered as 96 kHz / 24 bit WAV files.

"Robot Voices" is a downloadable sound collection of robotic speech. 50 Robots, many unique expressions per robot, delivered as 96 kHz / 24 bit WAV files.

"Robot Voices" is a downloadable sound collection of robotic speech. 50 Robots, many unique expressions per robot, delivered as 96 kHz / 24 bit WAV files.

"Robot Voices" is a downloadable sound collection of robotic speech. 50 Robots, many unique expressions per robot, delivered as 96 kHz / 24 bit WAV files.

"Robot Voices" is a downloadable sound collection of robotic speech. 50 Robots, many unique expressions per robot, delivered as 96 kHz / 24 bit WAV files.

Alibaba has open-sourced Qwen3-TTS, a text-to-speech model that clones voices from 3-second samples with 97ms latency across 10 languages.

Alibaba unveiled a triple threat in late December 2025: Fun-Audio-Chat-8B, an open-source speech-to-speech model with emotion-aware capabilities; Qwen-Image-Edit-2511, a major upgrade for consistent image editing; and VoiceDesign-VD-Flash, a controllable text-to-speech system that takes natural language instructions. These releases mark a significant shift in how developers can access frontier-level AI capabilities without proprietary restrictions, challenging closed-source models from OpenAI and Google.

kun432さんのスクラップ

Open-source text-to-speech powered by QWEN3-TTS **→ The app is already online. You can try it at anyspeak.ai and check repo on anyspeak-ai · GitHub ** Features * **TTS:** Multiple voices and Qwen3-TTS models (1.7B CustomVoice, VoiceDesign, Base). Supported languages: Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian, Spanish. * **Speaker tags:** Use `[SPEAKER_NAME]` in text for multi-speaker output. * **Mood tags:** Use `[SPEAKER:mood]` (e.g. `[VIVIAN:happy]`) for emotional expressiveness. * **Timing tags:** Use `[SPEAKER:mood:timing]` (e.g. `[RYAN:+1.5]` for 1.5 s pause after, `[VIVIAN:happy:-0.3]` for shorter gap or even speaker overlapping) or the + TIMING button to control pacing between segments. * **Custom voices:** Voice Design by selecting attributes (gender, language, old/young, slow/fast, high/low, loud/soft, warm/rough). * **CLEAN:** Analyze and clean text for better TTS output. * **CHECK:** Quality check with Whisper (compare original vs. synthesized). * **IMPROVE:** Post-process generated audio by chunk: selectively regenerate mispronounced segments, trim silence, add pauses, or edit text and re-generate. Compare original vs. improved before applying. You can also edit finished MP3s afterward (e.g. different wording or different speaker at a specific place) without regenerating everything. * **VIDEO:** Create video from audio and images (with optional subtitles). * **Load & Save (MP3):** Load and save via MP3. Very powerful: you can continue working at the same place, and files store the original text with tags in their metadata. * **Import:** Load text from URL or file. * Run local - easy installation