| # Set bash to 'debug' mode, it will exit on : | |
| # -e 'error', -u 'undefined variable', -o ... 'error in pipeline', -x 'print commands', | |
| set -e | |
| set -u | |
| set -o pipefail | |
| # start command: | |
| # bash run_wavlm_smollm.sh --stage 8 --stop_stage 8 --ngpu 8 --tag tts_smollm_whole_rinit --train_args "--codec_token_per_frame 8 --codec_token_in_use 8 --loss_region whole" | |
| train_jsons="" | |
| valid_jsons="" | |
| test_jsons="" | |
| # TTS | |
| train_jsons+="dump_wavlm/raw_tts_librispeech/train_960/data.json dump_wavlm/raw_tts_mls_en/mls_en_train/data.json" | |
| valid_jsons+="dump_wavlm/raw_tts_librispeech/dev/data.json " | |
| # Test sets | |
| test_jsons+="dump_wavlm/raw_codec_ssl_asr_librispeech/test_clean/data.json " | |
| test_jsons+="dump_wavlm/raw_codec_ssl_tts_librispeech/test_clean/data.json " | |
| train_config=conf/train_multiscale_smollm_360m.yaml | |
| inference_config=conf/decode_asr.yaml | |
| token_list_dir=data/token_list/llm_vocab # use lllm vocab | |
| bpe_opts="--subword_choice huggingface --subword_model HuggingFaceTB/SmolLM-1.7B" | |
| codec_opts="--codec_choice inhouse" | |
| ./speechlm.sh \ | |
| --skip_data_prep true \ | |
| --data_combo_name ls960_mlsen \ | |
| --fs 16000 \ | |
| --ngpu 4 \ | |
| --nj 16 \ | |
| --inference_nj 16 \ | |
| --nbest 10 \ | |
| --gpu_inference true \ | |
| --g2p g2p_en_no_space \ | |
| --cleaner tacotron \ | |
| --token_list_dir ${token_list_dir} \ | |
| --train_config ${train_config} \ | |
| --inference_config ${inference_config} \ | |
| --audio_format "flac.ark" \ | |
| --train_jsons "${train_jsons}" \ | |
| --valid_jsons "${valid_jsons}" \ | |
| --test_jsons "${test_jsons}" \ | |
| --dumpdir dump_wavlm \ | |
| ${bpe_opts} ${codec_opts} \ | |
| "$@" | |