FunCodec icon indicating copy to clipboard operation
FunCodec copied to clipboard

FunCodec is a research-oriented toolkit for audio quantization and downstream applications, such as text-to-speech synthesis, music generation et.al.

Results 27 FunCodec issues
Sort by recently updated
recently updated
newest added

您好,我想请问一下,encodec原论文中,量化器后可选的LM模型,应该如何训练和应用,是需要在encodec模型的config文件中进行一系列配置,还是需要重新写一个train文件呢?如果需要重新写一个train文件的话,应该怎么写呢?非常期待您的解答!

bash demo.sh --stage 2 --model_name ${model_name} --output_dir results --text "你好" \ --prompt_text "one of these is context" --prompt_audio "demo/8230_279154_000013_000003.wav" not support, self.phoneme_tokenizer uses g2p_en to convert english word into phn,is...

您好,当我在尝试多个gpu上训练时会遇到nccl库没有libnccl-net.so的报错,我已经确定我安装了版本为2.19.3的nccl库。而当我用单个gpu的训练时会遇到core dumped的问题。我目前用的是train-other-500 dev-other test-other这些数据集,请问是否对于这些报错有头绪

Huggingface下载的model.pth md5sum是4285b02283940535f4636787dc801dda,Modelscope下载的model.pth md5sum是277ad4e9033a825d78622d691df47b82,对同一音频进行编码得到的code也不相同,不是同一模型吗?

模型可以做到流式推理吗?

请问是否支持多机多卡呢,run.sh里有只支持了单机多卡配置

Should the audio be normed to [-1, 1]?

When i run stage 1 ``` bash encoding_decoding.sh --stage 1 --batch_size 16 --num_workers 4 --gpu_devices "0" \ --model_dir exp/${model_name} --bit_width 16000 \ --wav_scp input_wav.scp --out_dir outputs/codecs/ ``` It seems that...