4 ms·
Spark-TTS: Text-2-Speech Model Single-Stream Decoupled Tokens [pdf]
- mike978 2y agohttps://github.com/SparkAudio/Spark-TTS https://github.com/SparkAudio/Spark-TTS https://spark-tts.github.io/ https://spark-tts.github.io/
- smusamashah 2y agoThe voices with Chinese origin when generated as English samples do sound like a Chinese person speaking English. It is very interesting.
- fdafds 2y ago[flagged]
- vessenes 2y agoThis is really quite good at sounding like Donald, especially for the first half of the audio. I’ll probably play around with this for a bit; it’s. It clear to me how much variation you can get in voice in latent space. Anyway it looks to be a very high quality (at least) short form tts engine with open weights so thanks team!
- deknos 2y agoIs this really free software? I am really looking for _GOOD_ TTS software which is maintainable, really opensource (for every usage) and can do english/german/spanish/french/russian.
- popalchemist 2y agoZonos TTS is the SOTA, fully open-source (Apache license), and supports English, Japanese, Chinese, French, and German out of the box. You could train to add Russian, or run the output of this TTS through Meta's Seamless translation. https://github.com/Zyphra/Zonos https://github.com/Zyphra/Zonos
- maxglute 2y ago>Fast: our model runs with a real-time factor of ~2x on an RTX 4090 (i.e. generates 2 seconds of audio per 1 second of compute time) This is great as heavy TTS user. Waiting for real time 3x.