n4ze3m Releases tiny-voca, a 4.5M-Parameter English TTS Model Trained for $9.56 of GPU Time
Summary
Independent developer n4ze3m released tiny-voca, a 4.5M-parameter English text-to-speech model trained on 17.6 hours of synthetic speech for $9.56 of GPU time. It runs ~70x real time on four CPU threads, outputs 24kHz audio at up to 32s per utterance, and ranks 8th (M2 voice) and 13th (F2 voice) on the Open TTS Leaderboard with sub-0.8% WER. Apache-2.0, runs on CPU/WebAssembly/mobile.
Originally reported by huggingface.co
Read the original article →Original headline: n4ze3m Releases tiny-voca, a 4.5M-Parameter English TTS Model Trained for $9.56 of GPU Time