Skip to content

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

Dev gzf (#1475)

zhifu gao68f0603b10d1
diff --git a/.github/ISSUE_TEMPLATE/ask_questions.md b/.github/ISSUE_TEMPLATE/ask_questions.md
index 23a5a7f0b..4716faa9a 100644
--- a/.github/ISSUE_TEMPLATE/ask_questions.md
+++ b/.github/ISSUE_TEMPLATE/ask_questions.md
@@ -4,8 +4,12 @@ about: If you have questions, please first search existing issues and docs
labels: 'question, needs triage'
---
+Notice: In order to resolve issues more efficiently, please raise issue following the template.
+(注意:为了更加高效率解决您遇到的问题,请按照模板提问,补充细节)
+
## ❓ Questions and Help
+
### Before asking:
1. search the issues.
2. search the docs.
diff --git a/.github/ISSUE_TEMPLATE/bug_report.md b/.github/ISSUE_TEMPLATE/bug_report.md
index 82ba025fc..80b03179a 100644
--- a/.github/ISSUE_TEMPLATE/bug_report.md
+++ b/.github/ISSUE_TEMPLATE/bug_report.md
@@ -4,6 +4,9 @@ about: Submit a bug report to help us improve
labels: 'bug, needs triage'
---
+Notice: In order to resolve issues more efficiently, please raise issue following the template.
+(注意:为了更加高效率解决您遇到的问题,请按照模板提问,补充细节)
+
## 🐛 Bug
<!-- A clear and concise description of what the bug is. -->
diff --git a/examples/industrial_data_pretraining/llm_asr/conf/whisper_qwen_linear.yaml b/examples/industrial_data_pretraining/llm_asr/conf/whisper_qwen_linear.yaml
index 4603f805b..fb00b122c 100644
--- a/examples/industrial_data_pretraining/llm_asr/conf/whisper_qwen_linear.yaml
+++ b/examples/industrial_data_pretraining/llm_asr/conf/whisper_qwen_linear.yaml
@@ -12,17 +12,16 @@ model_conf:
length_normalized_loss: true
# encoder
-audio_encoder: iic/Whisper-large-v2 #iic/Whisper-large-v3
+audio_encoder: "/nfs/zhifu.gzf/init_model/Whisper-large-v3" #iic/Whisper-large-v3
audio_encoder_conf:
hub: ms
freeze: true
- init_param_path: "/nfs/maziyang.mzy/models/vicuna-7b-v1.5"
-llm: Vicuna
+llm: Qwen1.5-7b-chat
llm_conf:
hub: hf
freeze: true
- init_param_path: "/nfs/maziyang.mzy/models/vicuna-7b-v1.5"
+ init_param_path: "/nfs/zhifu.gzf/init_model/qwen/Qwen1___5-7B-Chat"
audio_adaptor: Linear
audio_adaptor_conf:
@@ -34,7 +33,7 @@ audio_adaptor_conf:
frontend: WhisperFrontend
frontend_conf:
fs: 16000
- whisper_model: large-v2
+ whisper_model: large-v3
do_pad_trim: true
permute: true # true: [bs, frames, dims]; false: [bs, dims, frames]
@@ -66,8 +65,9 @@ train_conf:
optim: adamw
optim_conf:
lr: 0.0001
- weight_decay: 0.000001
-scheduler: warmuplr
+ weight_decay: 0.000000
+
+scheduler: custom_lambdalr
scheduler_conf:
warmup_steps: 1000
@@ -83,12 +83,12 @@ dataset_conf:
preprocessor_text: TextPreprocessRemovePunctuation
audio_adaptor_downsample_rate: ${audio_adaptor_conf.downsample_rate}
audio_encoder_downsample_rate: 2
- prompt: "<|startoftranscription|><|zh|><|transcribe|><|zh|><|notimestamps|><|wo_itn|>"
+# prompt: "<|startoftranscription|><|zh|><|transcribe|><|zh|><|notimestamps|><|wo_itn|>"
tokenizer: HuggingfaceTokenizer
tokenizer_conf:
unk_symbol: <unk>
- init_param_path: "/nfs/maziyang.mzy/models/vicuna-7b-v1.5"
+ init_param_path: "/nfs/zhifu.gzf/init_model/qwen/Qwen1___5-7B-Chat"
diff --git a/funasr/datasets/audio_datasets/index_ds.py b/funasr/datasets/audio_datasets/index_ds.py
index 12ffd23d5..34f7b4fa6 100644
--- a/funasr/datasets/audio_datasets/index_ds.py
+++ b/funasr/datasets/audio_datasets/index_ds.py
@@ -99,7 +99,8 @@ class IndexDSJsonlRankFull(torch.utils.data.Dataset):
target = data["target"]
source_len = data.get("source_len", 1)
target_len = data.get("target_len", 0)
-
+ if "aishell" in source:
+ target = target.replace(" ", "")
contents.append({"source": source,
"prompt": prompt,
"target": target,