FunASR

阿里巴巴语音识别funasr的android部署
modelscope-funasr本地部署到安卓FunASR
funasr-android
UniASR语音识别-闽南语-通用-16k

sherpa-onnx
Releases/asr-models
Releases/tts-models
speech-enhancement-models
speaker-recongition-models
speaker-segmentation-models
Text-to-speech (TTS)
Automatic Speech Recognition
Generate subtitles for videos
Speaker diarization
ModelScope csukuangfj

VITS
facebook/mms-tts-nan
csukuangfj/vits-mms-nan

TSukiLen/whisper-medium-chinese-tw-minnan
TSukiLen/whisper-small-chinese-tw-minnan

生成的音频

import torch
from transformers import VitsTokenizer, VitsModel, set_seed

tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-nan")
model = VitsModel.from_pretrained("facebook/mms-tts-nan")

inputs = tokenizer(text="wo shi shui", return_tensors="pt")

# set_seed(555)

with torch.no_grad():
   outputs = model(**inputs)

waveform = outputs.waveform[0]

保存

import torchaudio

# 保存为 WAV 文件（默认采样率 16kHz）
torchaudio.save("output.wav", waveform.unsqueeze(0), model.config.sampling_rate)