How to train a bilingual Chinese-English zipformer model for speech recognition

[choshiho]

trafficstars

According to the link shown below, https://huggingface.co/csukuangfj/sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20

How to train a bilingual Chinese-English zipformer model from icefall/egs/librispeech/ASR/pruned_transducer_stateless7 _streaming?

Because the data preparation part is based on English speech corpus LibriSpeech，there is not a data preparation and training procedure about bilingual corpus in that directory.

[csukuangfj]

1. Please follow https://github.com/k2-fsa/icefall/tree/master/egs/tal_csasr/ASR to prepare your data

2. Use https://github.com/k2-fsa/icefall/tree/master/egs/librispeech/ASR/pruned_transducer_stateless7_streaming to replace pruned_transducer_stateless5

[choshiho]

1. Please follow https://github.com/k2-fsa/icefall/tree/master/egs/tal_csasr/ASR to prepare your data
2. Use https://github.com/k2-fsa/icefall/tree/master/egs/librispeech/ASR/pruned_transducer_stateless7_streaming to replace pruned_transducer_stateless5

hi, according to your second reply message, the data preparation part is based on English speech corpus LibriSpeech，there is not a data preparation and training procedure about bilingual corpus in that directory.

[csukuangfj]

Please follow step 1 to prepare your data. Step 1 is for bilingual dataset.

[choshiho]

Please follow step 1 to prepare your data. Step 1 is for bilingual dataset.

One more question, there are several lines containing librispeech_cuts_*.jsonl.gz in icefall/egs/librispeech/ASR/pruned_transducer_stateless7_streaming/asr_datamodule.py, should I replace them with data prepared according to your step 1?

[csukuangfj]

should I replace them with data prepared according to your step 1?

Yes, you are right.