Datasets:

MediaTek-Research
/

TASTE-Dump

+accum_grad: 2
+batch_idx: 4315
+checkpoint: /proj/mtklmadm/dev/mtk53678/rtslm_storage/code/rtslm/CosyVoice/examples/emilia/taste/exp/llm/torch_ddp/stage2/0127_stg2_eos-drop_rvq-d256-l4-k512_sum/checkpoint_best.pt
+config: /proj/mtklmadm/dev/mtk53678/rtslm_storage/code/rtslm/CosyVoice/examples/emilia/taste/conf/stage2/0207_stg2_eos_rvq-d256-l4-k512_sum_smaller-lr.yaml
+cv_data: /proj/mtklmadm/dev/mtk53678/rtslm_storage/code/rtslm/CosyVoice/examples/emilia/taste/data/dev.data.list
+deepscale: false
+deepscale_config: null
+deepspeed: false
+deepspeed_config: ./conf/customized_ds.json
+dist_backend: nccl
+dtype: fp32
+epoch: 0
+grad_clip: 5
+grad_norm: !!python/object/apply:torch._utils._rebuild_tensor_v2
+- !!python/object/apply:torch.storage._load_from_bytes
+  - !!binary |
+    gAKKCmz8nEb5IGqoUBkugAJN6QMugAJ9cQAoWBAAAABwcm90b2NvbF92ZXJzaW9ucQFN6QNYDQAA
+    AGxpdHRsZV9lbmRpYW5xAohYCgAAAHR5cGVfc2l6ZXNxA31xBChYBQAAAHNob3J0cQVLAlgDAAAA
+    aW50cQZLBFgEAAAAbG9uZ3EHSwR1dS6AAihYBwAAAHN0b3JhZ2VxAGN0b3JjaApGbG9hdFN0b3Jh
+    Z2UKcQFYCgAAADQ5MjUwMTE1ODRxAlgGAAAAY3VkYTowcQNLAU50cQRRLoACXXEAWAoAAAA0OTI1
+    MDExNTg0cQFhLgEAAAAAAAAAG8EsPw==
+- 0
+- !!python/tuple []
+- !!python/tuple []
+- false
+- !!python/object/apply:collections.OrderedDict
+  - []
+log_interval: 100
+loss_dict:
+  acc: 0.3778303575254848
+  len: 0.0
+  loss: 2.296471059830398
+lr: 0.00014604717857689557
+max_epoch: 3
+model: llm
+model_dir: /proj/mtklmadm/dev/mtk53678/rtslm/CosyVoice/examples/emilia/taste/exp/llm/torch_ddp/stage2/0207_stg2_eos_rvq-d256-l4-k512_sum_smaller-lr
+num_workers: 6
+optim: adam
+optim_conf:
+  lr: 0.00016
+pin_memory: true
+prefetch: 64
+save_per_step: 2000
+save_states: model+optimizer
+save_time: 08/02/2025 20:41:10
+scheduler: warmuplr
+scheduler_conf:
+  warmup_steps: 5000
+step: 5999
+tag: CV
+tensorboard_dir: /proj/mtklmadm/dev/mtk53678/rtslm/CosyVoice/examples/emilia/taste/tensorboard/llm/torch_ddp/stage2/0207_stg2_eos_rvq-d256-l4-k512_sum_smaller-lr
+timeout: 30
+train_data: /proj/mtklmadm/dev/mtk53678/rtslm_storage/code/rtslm/CosyVoice/examples/emilia/taste/data/train_with_libritts.data.list
+train_engine: torch_ddp

checkpoints/taste/config.yaml ADDED Viewed

	@@ -0,0 +1,312 @@

+# set random seed, so that you may reproduce your result.
+__set_seed1: !apply:random.seed [1986]
+__set_seed2: !apply:numpy.random.seed [1986]
+__set_seed3: !apply:torch.manual_seed [1986]
+__set_seed4: !apply:torch.cuda.manual_seed_all [1986]
+# fixed params
+sample_rate: 22050
+text_encoder_input_size: 512
+llm_input_size: 1024
+llm_output_size: 1024
+spk_embed_dim: 192
+## for audio branch
+is_word_level: True
+add_eos: True
+## add drop eos before llm functionality
+drop_eos_before_llm: False
+audio_embed_dim: 1280
+audio_token_encoder_input_size: 512
+whisper_tokenizer_fpath: /path/to/your/RTSLM_STORAGE_DIR/pretrained_models/distil-whisper-large-v3 # requires modification
+# training & dataloading commonly used params
+max_frames_in_batch: 16000 # original: 16000 (bsz~48)
+accum_grad: 2
+# model params
+# for all class/function included in this repo, we use !<name> or !<new> for intialization, so that user may find all corresponding class/function according to one single yaml.
+# for system/third_party class/function, we do not require this.
+llm: !new:cosyvoice.llm.audio_llm.TransformerJointLM
+    load_partial_list: [] # fully loaded
+        # 'text_encoder',
+        # 'llm',
+        # 'llm_decoder',
+        # 'audio_tokenizer', # add this would probably simulate fully loaded
+        # 'audio_token_encoder',
+        # 'audio_embed_affine_layer',
+    freeze_partial_list: [
+        'audio_tokenizer.audio_joint_encoder_segmenter.audio_encoder',
+        'audio_tokenizer.audio_joint_encoder_segmenter.audio_segmenter.decoder.embed',
+        'audio_tokenizer.audio_joint_encoder_segmenter.audio_segmenter.decoder.layers.0',
+    ]
+    freeze_partial_during_init: True
+    text_encoder_input_size: !ref <text_encoder_input_size>
+    llm_input_size: !ref <llm_input_size>
+    llm_output_size: !ref <llm_output_size>
+    text_token_size: 51866
+    speech_token_size: 4096
+    length_normalized_loss: True
+    lsm_weight: 0
+    spk_embed_dim: !ref <spk_embed_dim>
+    # for audio branch
+    audio_embed_dim: !ref <audio_embed_dim>
+    audio_token_encoder_input_size: !ref <audio_token_encoder_input_size>
+    is_word_level: !ref <is_word_level>
+    drop_eos_before_llm: !ref <drop_eos_before_llm>
+    fuse_encoded_audio_text_type: 'weighted_sum'
+    fuse_encoded_audio_text_kwargs:
+        normalize: False
+        use_trainable_weight: True
+        use_layer_norm: False
+        weight_init_type: 'balance'
+    audio_tokenizer: !new:cosyvoice.audio.audio_tokenizer.JointEncoderSegmenterAudioTokenizer
+        audio_joint_encoder_segmenter: !new:cosyvoice.audio.audio_joint_encoder_segmenter.WhisperAudioJointEncoderSegmenter
+            model_name_or_path: !ref <whisper_tokenizer_fpath>
+            attn_implementation: eager
+            # attn_implementation: flash_attention_2
+            # dtype: bfloat16
+            dtype: float32
+            use_custom: True
+            forward_type: asr_attn_pooling
+            make_v_proj_identity: True
+            is_word_level: !ref <is_word_level>
+            skip_prefix_idx: 4   # skip bos (0), task_id (1), lang_id (2), no-timestamp (3), start from idx=4
+            # skip_postfix_idx: -1 # skip eos (-1), currently unsupported
+        audio_quantizer: !new:cosyvoice.audio.audio_quantizer.RVQAudioQuantizer
+            dim: !ref <audio_embed_dim>
+            num_quantizers: 4
+            codebook_dim: 256
+            codebook_size: 512
+            decay: 0.99
+            quantize_dropout: True
+            kmeans_init: True
+    # audio token encoder is parallel with the text_encoder below
+    audio_token_encoder: !new:cosyvoice.transformer.encoder.ConformerEncoder
+        input_size: !ref <audio_token_encoder_input_size>
+        output_size: 1024
+        attention_heads: 8
+        linear_units: 2048
+        num_blocks: 2
+        dropout_rate: 0.1
+        positional_dropout_rate: 0.1
+        attention_dropout_rate: 0
+        normalize_before: True
+        input_layer: 'linear'
+        pos_enc_layer_type: 'rel_pos_espnet'
+        selfattention_layer_type: 'rel_selfattn'
+        use_cnn_module: False
+        macaron_style: False
+        use_dynamic_chunk: False
+        use_dynamic_left_chunk: False
+        static_chunk_size: 1
+    # end of audio branch
+    text_encoder: !new:cosyvoice.transformer.encoder.ConformerEncoder
+        input_size: !ref <text_encoder_input_size>
+        output_size: 1024
+        attention_heads: 8
+        linear_units: 2048
+        num_blocks: 3
+        dropout_rate: 0.1
+        positional_dropout_rate: 0.1
+        attention_dropout_rate: 0
+        normalize_before: True
+        input_layer: 'linear'
+        pos_enc_layer_type: 'rel_pos_espnet'
+        selfattention_layer_type: 'rel_selfattn'
+        use_cnn_module: False
+        macaron_style: False
+        use_dynamic_chunk: False
+        use_dynamic_left_chunk: False
+        static_chunk_size: 1
+    llm: !new:cosyvoice.transformer.encoder.TransformerEncoder
+        input_size: !ref <llm_input_size>
+        output_size: !ref <llm_output_size>
+        attention_heads: 8
+        linear_units: 2048
+        num_blocks: 7
+        dropout_rate: 0.1
+        positional_dropout_rate: 0.1
+        attention_dropout_rate: 0
+        input_layer: 'linear_legacy'
+        pos_enc_layer_type: 'rel_pos_espnet'
+        selfattention_layer_type: 'rel_selfattn'
+        static_chunk_size: 1
+flow: !new:cosyvoice.flow.flow.MaskedDiffWithXvec
+    input_size: 512
+    output_size: 80
+    spk_embed_dim: !ref <spk_embed_dim>
+    output_type: 'mel'
+    vocab_size: 4096
+    input_frame_rate: 50
+    only_mask_loss: True
+    encoder: !new:cosyvoice.transformer.encoder.ConformerEncoder
+        output_size: 512
+        attention_heads: 8
+        linear_units: 2048
+        num_blocks: 6
+        dropout_rate: 0.1
+        positional_dropout_rate: 0.1
+        attention_dropout_rate: 0.1
+        normalize_before: True
+        input_layer: 'linear'
+        pos_enc_layer_type: 'rel_pos_espnet'
+        selfattention_layer_type: 'rel_selfattn'
+        input_size: 512
+        use_cnn_module: False
+        macaron_style: False
+    length_regulator: !new:cosyvoice.flow.length_regulator.InterpolateRegulator
+        channels: 80
+        sampling_ratios: [1, 1, 1, 1]
+    decoder: !new:cosyvoice.flow.flow_matching.ConditionalCFM
+        in_channels: 240
+        n_spks: 1
+        spk_emb_dim: 80
+        cfm_params: !new:omegaconf.DictConfig
+            content:
+                sigma_min: 1e-06
+                solver: 'euler'
+                t_scheduler: 'cosine'
+                training_cfg_rate: 0.2
+                inference_cfg_rate: 0.7
+                reg_loss_type: 'l1'
+        estimator: !new:cosyvoice.flow.decoder.ConditionalDecoder
+            in_channels: 320
+            out_channels: 80
+            channels: [256, 256]
+            dropout: 0
+            attention_head_dim: 64
+            n_blocks: 4
+            num_mid_blocks: 12
+            num_heads: 8
+            act_fn: 'gelu'
+hift: !new:cosyvoice.hifigan.generator.HiFTGenerator
+    in_channels: 80
+    base_channels: 512
+    nb_harmonics: 8
+    sampling_rate: !ref <sample_rate>
+    nsf_alpha: 0.1
+    nsf_sigma: 0.003
+    nsf_voiced_threshold: 10
+    upsample_rates: [8, 8]
+    upsample_kernel_sizes: [16, 16]
+    istft_params:
+        n_fft: 16
+        hop_len: 4
+    resblock_kernel_sizes: [3, 7, 11]
+    resblock_dilation_sizes: [[1, 3, 5], [1, 3, 5], [1, 3, 5]]
+    source_resblock_kernel_sizes: [7, 11]
+    source_resblock_dilation_sizes: [[1, 3, 5], [1, 3, 5]]
+    lrelu_slope: 0.1
+    audio_limit: 0.99
+    f0_predictor: !new:cosyvoice.hifigan.f0_predictor.ConvRNNF0Predictor
+        num_class: 1
+        in_channels: 80
+        cond_channels: 512
+# processor functions
+parquet_opener: !name:cosyvoice.dataset.processor.parquet_opener
+get_tokenizer: !name:whisper.tokenizer.get_tokenizer
+    multilingual: True
+    num_languages: 100
+    language: 'en'
+    task: 'transcribe'
+allowed_special: 'all'
+tokenize: !name:cosyvoice.dataset.processor.tokenize_by_words
+    get_tokenizer: !ref <get_tokenizer>
+    allowed_special: !ref <allowed_special>
+    use_asr_text: True
+tokenize_whisper: !name:cosyvoice.dataset.processor.tokenize_whisper
+    whisper_tokenizer_name_or_fpath: !ref <whisper_tokenizer_fpath>
+    task: 'transcribe'
+    language: 'en'
+    no_timestamps: True
+    add_bos: True
+    add_eos: !ref <add_eos>
+    use_asr_text: True
+    overwrite_text_token: True
+    use_wrapped: True
+filter: !name:cosyvoice.dataset.processor.filter
+    max_length: 40960
+    min_length: 0
+    token_max_length: 200
+    token_min_length: 1
+resample: !name:cosyvoice.dataset.processor.resample
+    resample_rate: !ref <sample_rate>
+feat_extractor: !name:matcha.utils.audio.mel_spectrogram
+    n_fft: 1024
+    num_mels: 80
+    sampling_rate: !ref <sample_rate>
+    hop_size: 256
+    win_size: 1024
+    fmin: 0
+    fmax: 8000
+    center: False
+# audio_extractor: !new:cosyvoice.audio.audio_extractor.WhisperAudioExtractor
+#     model_name_or_path: "/proj/mtklmadm/dev/mtk53678/rtslm_storage/pretrained_models/whisper-large-v3"
+#     pad_to_whisper_input_size: False
+#     use_orig_length: True
+audio_extractor: !new:funasr.frontends.whisper_frontend.WhisperFrontend
+    whisper_model: large-v3
+    do_pad_trim: True
+    permute: True
+# audio_encoder: !new:cosyvoice.audio.audio_encoder.SenseVoiceAudioEncoder
+#     model_card: "/proj/mtklmadm/dev/mtk53678/rtslm_storage/pretrained_models/SenseVoiceSmall"
+#     model_code_dir: "/proj/mtklmadm/dev/mtk53678/rtslm/CosyVoice/cosyvoice/audio/customized_sensevoice/model.py"
+#     hub: "ms"
+# audio_extractor: !ref <audio_encoder.frontend>
+extract_audio: !name:cosyvoice.dataset.processor.extract_audio
+    # get_audio_extractor: !ref <get_audio_extractor>
+    audio_extractor: !ref <audio_extractor>
+    target_sample_rate: 16_000
+    # Maybe we should add prepending sensevoice tokens here.
+compute_fbank: !name:cosyvoice.dataset.processor.compute_fbank
+    feat_extractor: !ref <feat_extractor>
+parse_embedding: !name:cosyvoice.dataset.processor.parse_embedding
+    normalize: True
+shuffle: !name:cosyvoice.dataset.processor.shuffle
+    shuffle_size: 1000
+sort: !name:cosyvoice.dataset.processor.sort
+    sort_size: 500  # sort_size should be less than shuffle_size
+batch: !name:cosyvoice.dataset.processor.batch
+    batch_type: 'dynamic'
+    max_frames_in_batch: !ref <max_frames_in_batch> # 4800
+padding: !name:cosyvoice.dataset.processor.padding
+    use_spk_embedding: False # change to True during sft
+    has_audio_branch: True # add audio branch related padding
+    use_asr_text: False # avoid directly using asr text and alignments for compatibility
+    use_auto_audio_len: False
+    requires_words_index: !ref <is_word_level> # for word-level audio token
+# dataset processor pipeline
+data_pipeline: [
+    !ref <parquet_opener>,
+    !ref <tokenize>, # using tokenize_whisper can skip this
+    !ref <tokenize_whisper>,
+    !ref <filter>,
+    !ref <extract_audio>, # added for audio branch, NOTE: should be in front of resample and compute fbank!!
+    !ref <resample>,
+    !ref <compute_fbank>,
+    !ref <parse_embedding>,
+    !ref <shuffle>,
+    !ref <sort>,
+    !ref <batch>,
+    !ref <padding>,
+]
+# train conf
+train_conf:
+    optim: adam
+    optim_conf:
+        lr: 0.00016 # change to 0.001 if you want to train flow from scratch
+    scheduler: warmuplr
+    scheduler_conf:
+        # warmup_steps: 25000
+        warmup_steps: 5000
+    # max_epoch: 200
+    max_epoch: 3
+    grad_clip: 5
+    accum_grad: !ref <accum_grad> # 2
+    log_interval: 100
+    save_per_step: 2000

checkpoints/text-only_baseline/checkpoint_best.pt ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:e75ab1a4d8c6592de30b4ee9add5225056faeff2ab78f03c9a19d3d57b38ae23
+size 529227076

checkpoints/text-only_baseline/checkpoint_best.yaml ADDED Viewed

	@@ -0,0 +1,54 @@

+accum_grad: 2
+batch_idx: 2322
+checkpoint: /proj/mtklmadm/dev/mtk53678/rtslm_storage/code/rtslm/CosyVoice/examples/emilia/taste/exp/llm/torch_ddp/text-only_baseline/checkpoint_best.pt
+config: /proj/mtklmadm/dev/mtk53678/rtslm_storage/code/rtslm/CosyVoice/examples/emilia/taste/conf/text-only_baseline_contd.yaml
+cv_data: /proj/mtklmadm/dev/mtk53678/rtslm_storage/code/rtslm/CosyVoice/examples/emilia/taste/data/dev.data.list
+deepscale: false
+deepscale_config: null
+deepspeed: false
+deepspeed_config: ./conf/customized_ds.json
+dist_backend: nccl
+dtype: fp32
+epoch: 1
+grad_clip: 5
+grad_norm: !!python/object/apply:torch._utils._rebuild_tensor_v2
+- !!python/object/apply:torch.storage._load_from_bytes
+  - !!binary |
+    gAKKCmz8nEb5IGqoUBkugAJN6QMugAJ9cQAoWBAAAABwcm90b2NvbF92ZXJzaW9ucQFN6QNYDQAA
+    AGxpdHRsZV9lbmRpYW5xAohYCgAAAHR5cGVfc2l6ZXNxA31xBChYBQAAAHNob3J0cQVLAlgDAAAA
+    aW50cQZLBFgEAAAAbG9uZ3EHSwR1dS6AAihYBwAAAHN0b3JhZ2VxAGN0b3JjaApGbG9hdFN0b3Jh
+    Z2UKcQFYCQAAADQ5NjU1NzI4MHECWAYAAABjdWRhOjBxA0sBTnRxBFEugAJdcQBYCQAAADQ5NjU1
+    NzI4MHEBYS4BAAAAAAAAACAE0T0=
+- 0
+- !!python/tuple []
+- !!python/tuple []
+- false
+- !!python/object/apply:collections.OrderedDict
+  - []
+log_interval: 100
+loss_dict:
+  acc: 0.3084577530809149
+  len: 0.0
+  loss: 2.7354950856894007
+lr: 5.929224290931882e-05
+max_epoch: 2
+model: llm
+model_dir: /proj/mtklmadm/dev/mtk53678/rtslm/CosyVoice/examples/emilia/taste/exp/llm/torch_ddp/text-only_baseline_contd
+num_workers: 4
+optim: adam
+optim_conf:
+  lr: 0.00015
+pin_memory: true
+prefetch: 128
+save_per_step: 4000
+save_states: model+optimizer
+save_time: 10/02/2025 09:14:22
+scheduler: warmuplr
+scheduler_conf:
+  warmup_steps: 10000
+step: 63999
+tag: CV
+tensorboard_dir: /proj/mtklmadm/dev/mtk53678/rtslm/CosyVoice/examples/emilia/taste/tensorboard/llm/torch_ddp/text-only_baseline_contd
+timeout: 30
+train_data: /proj/mtklmadm/dev/mtk53678/rtslm_storage/code/rtslm/CosyVoice/examples/emilia/taste/data/train_with_libritts.data.list
+train_engine: torch_ddp

checkpoints/text-only_baseline/config.yaml ADDED Viewed

	@@ -0,0 +1,214 @@

+# set random seed, so that you may reproduce your result.
+__set_seed1: !apply:random.seed [1986]
+__set_seed2: !apply:numpy.random.seed [1986]
+__set_seed3: !apply:torch.manual_seed [1986]
+__set_seed4: !apply:torch.cuda.manual_seed_all [1986]
+# fixed params
+sample_rate: 22050
+text_encoder_input_size: 512
+llm_input_size: 1024
+llm_output_size: 1024
+spk_embed_dim: 192
+# others
+# add_eos: True
+whisper_tokenizer_fpath: /path/to/your/RTSLM_STORAGE_DIR/pretrained_models/distil-whisper-large-v3 # requires modification
+# model params
+# for all class/function included in this repo, we use !<name> or !<new> for intialization, so that user may find all corresponding class/function according to one single yaml.
+# for system/third_party class/function, we do not require this.
+llm: !new:cosyvoice.llm.llm.TransformerLM
+    text_encoder_input_size: !ref <text_encoder_input_size>
+    llm_input_size: !ref <llm_input_size>
+    llm_output_size: !ref <llm_output_size>
+    text_token_size: 51866
+    speech_token_size: 4096
+    length_normalized_loss: True
+    lsm_weight: 0
+    spk_embed_dim: !ref <spk_embed_dim>
+    text_encoder: !new:cosyvoice.transformer.encoder.ConformerEncoder
+        input_size: !ref <text_encoder_input_size>
+        output_size: 1024
+        attention_heads: 8
+        linear_units: 2048
+        num_blocks: 3
+        dropout_rate: 0.1
+        positional_dropout_rate: 0.1
+        attention_dropout_rate: 0
+        normalize_before: True
+        input_layer: 'linear'
+        pos_enc_layer_type: 'rel_pos_espnet'
+        selfattention_layer_type: 'rel_selfattn'
+        use_cnn_module: False
+        macaron_style: False
+        use_dynamic_chunk: False
+        use_dynamic_left_chunk: False
+        static_chunk_size: 1
+    llm: !new:cosyvoice.transformer.encoder.TransformerEncoder
+        input_size: !ref <llm_input_size>
+        output_size: !ref <llm_output_size>
+        attention_heads: 8
+        linear_units: 2048
+        num_blocks: 7
+        dropout_rate: 0.1
+        positional_dropout_rate: 0.1
+        attention_dropout_rate: 0
+        input_layer: 'linear_legacy'
+        pos_enc_layer_type: 'rel_pos_espnet'
+        selfattention_layer_type: 'rel_selfattn'
+        static_chunk_size: 1
+flow: !new:cosyvoice.flow.flow.MaskedDiffWithXvec
+    input_size: 512
+    output_size: 80
+    spk_embed_dim: !ref <spk_embed_dim>
+    output_type: 'mel'
+    vocab_size: 4096
+    input_frame_rate: 50
+    only_mask_loss: True
+    encoder: !new:cosyvoice.transformer.encoder.ConformerEncoder
+        output_size: 512
+        attention_heads: 8
+        linear_units: 2048
+        num_blocks: 6
+        dropout_rate: 0.1
+        positional_dropout_rate: 0.1
+        attention_dropout_rate: 0.1
+        normalize_before: True
+        input_layer: 'linear'
+        pos_enc_layer_type: 'rel_pos_espnet'
+        selfattention_layer_type: 'rel_selfattn'
+        input_size: 512
+        use_cnn_module: False
+        macaron_style: False
+    length_regulator: !new:cosyvoice.flow.length_regulator.InterpolateRegulator
+        channels: 80
+        sampling_ratios: [1, 1, 1, 1]
+    decoder: !new:cosyvoice.flow.flow_matching.ConditionalCFM
+        in_channels: 240
+        n_spks: 1
+        spk_emb_dim: 80
+        cfm_params: !new:omegaconf.DictConfig
+            content:
+                sigma_min: 1e-06
+                solver: 'euler'
+                t_scheduler: 'cosine'
+                training_cfg_rate: 0.2
+                inference_cfg_rate: 0.7
+                reg_loss_type: 'l1'
+        estimator: !new:cosyvoice.flow.decoder.ConditionalDecoder
+            in_channels: 320
+            out_channels: 80
+            channels: [256, 256]
+            dropout: 0
+            attention_head_dim: 64
+            n_blocks: 4
+            num_mid_blocks: 12
+            num_heads: 8
+            act_fn: 'gelu'
+hift: !new:cosyvoice.hifigan.generator.HiFTGenerator
+    in_channels: 80
+    base_channels: 512
+    nb_harmonics: 8
+    sampling_rate: !ref <sample_rate>
+    nsf_alpha: 0.1
+    nsf_sigma: 0.003
+    nsf_voiced_threshold: 10
+    upsample_rates: [8, 8]
+    upsample_kernel_sizes: [16, 16]
+    istft_params:
+        n_fft: 16
+        hop_len: 4
+    resblock_kernel_sizes: [3, 7, 11]
+    resblock_dilation_sizes: [[1, 3, 5], [1, 3, 5], [1, 3, 5]]
+    source_resblock_kernel_sizes: [7, 11]
+    source_resblock_dilation_sizes: [[1, 3, 5], [1, 3, 5]]
+    lrelu_slope: 0.1
+    audio_limit: 0.99
+    f0_predictor: !new:cosyvoice.hifigan.f0_predictor.ConvRNNF0Predictor
+        num_class: 1
+        in_channels: 80
+        cond_channels: 512
+# processor functions
+parquet_opener: !name:cosyvoice.dataset.processor.parquet_opener
+get_tokenizer: !name:whisper.tokenizer.get_tokenizer
+    multilingual: True
+    num_languages: 100
+    language: 'en'
+    task: 'transcribe'
+allowed_special: 'all'
+tokenize: !name:cosyvoice.dataset.processor.tokenize_by_words
+    get_tokenizer: !ref <get_tokenizer>
+    allowed_special: !ref <allowed_special>
+    use_asr_text: True
+# tokenize_whisper: !name:cosyvoice.dataset.processor.tokenize_whisper
+#     whisper_tokenizer_name_or_fpath: !ref <whisper_tokenizer_fpath>
+#     task: 'transcribe'
+#     language: 'en'
+#     no_timestamps: True
+#     add_bos: True
+#     add_eos: !ref <add_eos>
+#     use_asr_text: True
+#     overwrite_text_token: True
+#     use_wrapped: True
+filter: !name:cosyvoice.dataset.processor.filter
+    max_length: 40960
+    min_length: 0
+    token_max_length: 200
+    token_min_length: 1
+resample: !name:cosyvoice.dataset.processor.resample
+    resample_rate: !ref <sample_rate>
+feat_extractor: !name:matcha.utils.audio.mel_spectrogram
+    n_fft: 1024
+    num_mels: 80
+    sampling_rate: !ref <sample_rate>
+    hop_size: 256
+    win_size: 1024
+    fmin: 0
+    fmax: 8000
+    center: False
+compute_fbank: !name:cosyvoice.dataset.processor.compute_fbank
+    feat_extractor: !ref <feat_extractor>
+parse_embedding: !name:cosyvoice.dataset.processor.parse_embedding
+    normalize: True
+shuffle: !name:cosyvoice.dataset.processor.shuffle
+    shuffle_size: 1000
+sort: !name:cosyvoice.dataset.processor.sort
+    sort_size: 500  # sort_size should be less than shuffle_size
+batch: !name:cosyvoice.dataset.processor.batch
+    batch_type: 'dynamic'
+    max_frames_in_batch: 20000
+padding: !name:cosyvoice.dataset.processor.padding
+    use_spk_embedding: False # change to True during sft
+# dataset processor pipeline
+data_pipeline: [
+    !ref <parquet_opener>,
+    !ref <tokenize>,
+    # !ref <tokenize_whisper>,
+    !ref <filter>,
+    !ref <resample>,
+    !ref <compute_fbank>,
+    !ref <parse_embedding>,
+    !ref <shuffle>,
+    !ref <sort>,
+    !ref <batch>,
+    !ref <padding>,
+]
+# train conf
+train_conf:
+    optim: adam
+    optim_conf:
+        lr: 0.00015 # change to 0.001 if you want to train flow from scratch
+    scheduler: warmuplr
+    scheduler_conf:
+        warmup_steps: 10000
+    max_epoch: 2
+    grad_clip: 5
+    accum_grad: 2
+    log_interval: 100
+    save_per_step: 4000