Skip to content

Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.

update

游雁362a6de3c850
diff --git a/examples/aishell/branchformer/conf/branchformer_12e_6d_2048_256.yaml b/examples/aishell/branchformer/conf/branchformer_12e_6d_2048_256.yaml
index 09b1e0f4f..d6caf5d27 100644
--- a/examples/aishell/branchformer/conf/branchformer_12e_6d_2048_256.yaml
+++ b/examples/aishell/branchformer/conf/branchformer_12e_6d_2048_256.yaml
@@ -94,13 +94,16 @@ scheduler_conf:
dataset: AudioDataset
dataset_conf:
index_ds: IndexDSJsonl
- batch_sampler: DynamicBatchLocalShuffleSampler
+ batch_sampler: RankFullLocalShuffleBatchSampler
batch_type: example # example or length
- batch_size: 1 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
+ batch_size: 32 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
max_token_length: 2048 # filter samples if source_token_len+target_token_len > max_token_length,
- buffer_size: 500
+ buffer_size: 1024
shuffle: True
num_workers: 4
+ preprocessor_speech: SpeechPreprocessSpeedPerturb
+ preprocessor_speech_conf:
+ speed_perturb: [0.9, 1.0, 1.1]
tokenizer: CharTokenizer
tokenizer_conf:
diff --git a/examples/aishell/conformer/conf/conformer_12e_6d_2048_256.yaml b/examples/aishell/conformer/conf/conformer_12e_6d_2048_256.yaml
index 1197e8723..8909b1596 100644
--- a/examples/aishell/conformer/conf/conformer_12e_6d_2048_256.yaml
+++ b/examples/aishell/conformer/conf/conformer_12e_6d_2048_256.yaml
@@ -94,13 +94,16 @@ scheduler_conf:
dataset: AudioDataset
dataset_conf:
index_ds: IndexDSJsonl
- batch_sampler: DynamicBatchLocalShuffleSampler
+ batch_sampler: RankFullLocalShuffleBatchSampler
batch_type: example # example or length
- batch_size: 1 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
+ batch_size: 32 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
max_token_length: 2048 # filter samples if source_token_len+target_token_len > max_token_length,
- buffer_size: 500
+ buffer_size: 1024
shuffle: True
num_workers: 4
+ preprocessor_speech: SpeechPreprocessSpeedPerturb
+ preprocessor_speech_conf:
+ speed_perturb: [0.9, 1.0, 1.1]
tokenizer: CharTokenizer
tokenizer_conf:
diff --git a/examples/aishell/conformer/run.sh b/examples/aishell/conformer/run.sh
index 28e6f90a1..947cc02e3 100755
--- a/examples/aishell/conformer/run.sh
+++ b/examples/aishell/conformer/run.sh
@@ -109,9 +109,14 @@ if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
echo "log_file: ${log_file}"
gpu_num=$(echo CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
- torchrun \
- --nnodes 1 \
- --nproc_per_node ${gpu_num} \
+# torchrun \
+# --nnodes 1 \
+# --nproc_per_node ${gpu_num}
+ cmd="python"
+ if [ ${gpu_num} -gt 1 ];then
+ cmd="torchrun --nnodes 1 --nproc_per_node ${gpu_num}"
+ fi
+ ${cmd} \
../../../funasr/bin/train.py \
--config-path "${workspace}/conf" \
--config-name "${config}" \
diff --git a/examples/aishell/e_branchformer/conf/e_branchformer_12e_6d_2048_256.yaml b/examples/aishell/e_branchformer/conf/e_branchformer_12e_6d_2048_256.yaml
index dea580e52..e6cc1e6c5 100644
--- a/examples/aishell/e_branchformer/conf/e_branchformer_12e_6d_2048_256.yaml
+++ b/examples/aishell/e_branchformer/conf/e_branchformer_12e_6d_2048_256.yaml
@@ -94,13 +94,16 @@ scheduler_conf:
dataset: AudioDataset
dataset_conf:
index_ds: IndexDSJsonl
- batch_sampler: DynamicBatchLocalShuffleSampler
+ batch_sampler: RankFullLocalShuffleBatchSampler
batch_type: example # example or length
- batch_size: 1 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
+ batch_size: 32 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
max_token_length: 2048 # filter samples if source_token_len+target_token_len > max_token_length,
- buffer_size: 500
+ buffer_size: 1024
shuffle: True
num_workers: 4
+ preprocessor_speech: SpeechPreprocessSpeedPerturb
+ preprocessor_speech_conf:
+ speed_perturb: [0.9, 1.0, 1.1]
tokenizer: CharTokenizer
tokenizer_conf:
diff --git a/examples/aishell/transformer/transformer_12e_6d_2048_256.yaml b/examples/aishell/transformer/conf/transformer_12e_6d_2048_256.yaml
similarity index 86%
rename from examples/aishell/transformer/transformer_12e_6d_2048_256.yaml
rename to examples/aishell/transformer/conf/transformer_12e_6d_2048_256.yaml
index 87814dc3b..24a63903a 100644
--- a/examples/aishell/transformer/transformer_12e_6d_2048_256.yaml
+++ b/examples/aishell/transformer/conf/transformer_12e_6d_2048_256.yaml
@@ -88,13 +88,16 @@ scheduler_conf:
dataset: AudioDataset
dataset_conf:
index_ds: IndexDSJsonl
- batch_sampler: DynamicBatchLocalShuffleSampler
+ batch_sampler: RankFullLocalShuffleBatchSampler
batch_type: example # example or length
- batch_size: 1 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
+ batch_size: 32 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
max_token_length: 2048 # filter samples if source_token_len+target_token_len > max_token_length,
- buffer_size: 500
+ buffer_size: 1024
shuffle: True
- num_workers: 0
+ num_workers: 4
+ preprocessor_speech: SpeechPreprocessSpeedPerturb
+ preprocessor_speech_conf:
+ speed_perturb: [0.9, 1.0, 1.1]
tokenizer: CharTokenizer
tokenizer_conf: