speecht5_tts-pld-fil-ONNX

ONNX / Transformers.js export of sapinsapin/speecht5_tts-pld-fil, bundled with the microsoft/speecht5_hifigan vocoder.

import { pipeline } from '@huggingface/transformers';
const tts = await pipeline('text-to-speech', 'internetoftim/speecht5_tts-pld-fil-ONNX', { dtype: 'q8', device: 'wasm' });
const out = await tts('Kumusta ka?', { speaker_embeddings: SPEAKER_EMBEDDING_URL });

Requires an external speaker embedding (SpeechT5 architecture). spm_char.model is byte-identical to microsoft/speecht5_tts, so the upstream fast tokenizer.json is reused. q8 weights total ~180 MB.