add flow cache model

.gitattributes

@@ -59,3 +59,4 @@ flow.decoder.estimator.fp16.l20.plan filter=lfs diff=lfs merge=lfs -text
 flow.decoder.estimator.fp16.a10.plan filter=lfs diff=lfs merge=lfs -text
 flow.decoder.estimator.fp16.v100.plan filter=lfs diff=lfs merge=lfs -text
 flow.encoder.fp16.zip filter=lfs diff=lfs merge=lfs -text
+flow.cache.pt filter=lfs diff=lfs merge=lfs -text

cosyvoice2.yaml

@@ -10,6 +10,12 @@ llm_input_size: 896
 llm_output_size: 896
 spk_embed_dim: 192
 qwen_pretrain_path: ''
+token_frame_rate: 25
+token_mel_ratio: 2
+
+# stream related params
+chunk_size: 25 # streaming inference chunk size, in token
+num_decoding_left_chunks: 1 # streaming inference flow decoder left chunk size, <0 means use all left chunks
 
 # 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.
@@ -20,6 +26,7 @@ llm: !new:cosyvoice.llm.llm.Qwen2LM
     speech_token_size: 6561
     length_normalized_loss: True
     lsm_weight: 0
+    mix_ratio: [5, 15]
     llm: !new:cosyvoice.llm.llm.Qwen2Encoder
         pretrain_path: !ref <qwen_pretrain_path>
     sampling: !name:cosyvoice.utils.common.ras_sampling
@@ -34,9 +41,9 @@ flow: !new:cosyvoice.flow.flow.CausalMaskedDiffWithXvec
     spk_embed_dim: !ref <spk_embed_dim>
     output_type: 'mel'
     vocab_size: 6561
-    input_frame_rate: 25
+    input_frame_rate: !ref <token_frame_rate>
     only_mask_loss: True
-    token_mel_ratio: 2
+    token_mel_ratio: !ref <token_mel_ratio>
     pre_lookahead_len: 3
     encoder: !new:cosyvoice.transformer.upsample_encoder.UpsampleConformerEncoder
         output_size: 512
@@ -53,6 +60,7 @@ flow: !new:cosyvoice.flow.flow.CausalMaskedDiffWithXvec
         input_size: 512
         use_cnn_module: False
         macaron_style: False
+        static_chunk_size: !ref <chunk_size>
     decoder: !new:cosyvoice.flow.flow_matching.CausalConditionalCFM
         in_channels: 240
         n_spks: 1
@@ -65,10 +73,9 @@ flow: !new:cosyvoice.flow.flow.CausalMaskedDiffWithXvec
                 training_cfg_rate: 0.2
                 inference_cfg_rate: 0.7
                 reg_loss_type: 'l1'
-        estimator: !new:cosyvoice.flow.decoder.ConditionalDecoder
+        estimator: !new:cosyvoice.flow.decoder.CausalConditionalDecoder
             in_channels: 320
             out_channels: 80
-            causal: True
             channels: [256]
             dropout: 0.0
             attention_head_dim: 64
@@ -76,6 +83,8 @@ flow: !new:cosyvoice.flow.flow.CausalMaskedDiffWithXvec
             num_mid_blocks: 12
             num_heads: 8
             act_fn: 'gelu'
+            static_chunk_size: !ref <chunk_size> * <token_mel_ratio>
+            num_decoding_left_chunks: !ref <num_decoding_left_chunks>
 
 hift: !new:cosyvoice.hifigan.generator.HiFTGenerator
     in_channels: 80
@@ -101,6 +110,25 @@ hift: !new:cosyvoice.hifigan.generator.HiFTGenerator
         in_channels: 80
         cond_channels: 512
 
+# gan related module
+mel_spec_transform1: !name:matcha.utils.audio.mel_spectrogram
+    n_fft: 1920
+    num_mels: 80
+    sampling_rate: !ref <sample_rate>
+    hop_size: 480
+    win_size: 1920
+    fmin: 0
+    fmax: null
+    center: False
+hifigan: !new:cosyvoice.hifigan.hifigan.HiFiGan
+    generator: !ref <hift>
+    discriminator: !new:cosyvoice.hifigan.discriminator.MultipleDiscriminator
+        mpd: !new:matcha.hifigan.models.MultiPeriodDiscriminator
+        mrd: !new:cosyvoice.hifigan.discriminator.MultiResSpecDiscriminator
+    mel_spec_transform: [
+        !ref <mel_spec_transform1>
+    ]
+
 # processor functions
 parquet_opener: !name:cosyvoice.dataset.processor.parquet_opener
 get_tokenizer: !name:cosyvoice.tokenizer.tokenizer.get_qwen_tokenizer
@@ -112,11 +140,13 @@ tokenize: !name:cosyvoice.dataset.processor.tokenize
     allowed_special: !ref <allowed_special>
 filter: !name:cosyvoice.dataset.processor.filter
     max_length: 40960
-    min_length: 0
+    min_length: 100
     token_max_length: 200
     token_min_length: 1
 resample: !name:cosyvoice.dataset.processor.resample
     resample_rate: !ref <sample_rate>
+truncate: !name:cosyvoice.dataset.processor.truncate
+    truncate_length: 24480 # must be a multiplier of hop_size
 feat_extractor: !name:matcha.utils.audio.mel_spectrogram
     n_fft: 1920
     num_mels: 80
@@ -128,6 +158,9 @@ feat_extractor: !name:matcha.utils.audio.mel_spectrogram
     center: False
 compute_fbank: !name:cosyvoice.dataset.processor.compute_fbank
     feat_extractor: !ref <feat_extractor>
+compute_f0: !name:cosyvoice.dataset.processor.compute_f0
+    sample_rate: !ref <sample_rate>
+    hop_size: 480
 parse_embedding: !name:cosyvoice.dataset.processor.parse_embedding
     normalize: True
 shuffle: !name:cosyvoice.dataset.processor.shuffle
@@ -137,4 +170,64 @@ sort: !name:cosyvoice.dataset.processor.sort
 batch: !name:cosyvoice.dataset.processor.batch
     batch_type: 'dynamic'
     max_frames_in_batch: 2000
-padding: !name:cosyvoice.dataset.processor.padding
+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 <filter>,
+    !ref <resample>,
+    !ref <compute_fbank>,
+    !ref <parse_embedding>,
+    !ref <shuffle>,
+    !ref <sort>,
+    !ref <batch>,
+    !ref <padding>,
+]
+data_pipeline_gan: [
+    !ref <parquet_opener>,
+    !ref <tokenize>,
+    !ref <filter>,
+    !ref <resample>,
+    !ref <truncate>,
+    !ref <compute_fbank>,
+    !ref <compute_f0>,
+    !ref <parse_embedding>,
+    !ref <shuffle>,
+    !ref <sort>,
+    !ref <batch>,
+    !ref <padding>,
+]
+
+# llm flow train conf
+train_conf:
+    optim: adam
+    optim_conf:
+        lr: 1e-5 # change to 1e-5 during sft
+    scheduler: constantlr # change to constantlr during sft
+    scheduler_conf:
+        warmup_steps: 2500
+    max_epoch: 200
+    grad_clip: 5
+    accum_grad: 2
+    log_interval: 100
+    save_per_step: -1
+
+# gan train conf
+train_conf_gan:
+    optim: adam
+    optim_conf:
+        lr: 0.0002 # use small lr for gan training
+    scheduler: constantlr
+    optim_d: adam
+    optim_conf_d:
+        lr: 0.0002 # use small lr for gan training
+    scheduler_d: constantlr
+    max_epoch: 200
+    grad_clip: 5
+    accum_grad: 1 # in gan training, accum_grad must be 1
+    log_interval: 100
+    save_per_step: -1