Trim default Python dependencies

README.md

@@ -22,7 +22,7 @@ license: apache-2.0
 ### Audio backend
 
 ```
-sudo apt install libsnffile1
+sudo apt install libsndfile1
 ```
 
 ### Python
@@ -36,10 +36,16 @@ conda activate fireredasr
 pip install -r requirements.txt
 ```
 
-### 安装pyaxengine
+`requirements.txt` 只包含 AX650N AED 推理路径需要的运行时依赖。其他脚本按需安装：
+
+- `fireredasr_onnx.py`: `pip install onnxruntime`
+- `fireredasr/speech2text.py` 的 LLM 路径: `pip install transformers peft`
+- `fireredasr/utils/wer.py --do_tn 1`: `pip install cn2an`
+
+### 手动安装pyaxengine
 
 ```
-wget https://github.com/AXERA-TECH/pyaxengine/releases/download/0.1.3.rc1/axengine-0.1.3-py3-none-any.whl
+wget https://github.com/AXERA-TECH/pyaxengine/releases/download/0.1.3.rc2/axengine-0.1.3-py3-none-any.whl
 pip install axengine-0.1.3-py3-none-any.whl
 ```
 

fireredasr/models/fireredasr.py

@@ -5,9 +5,7 @@ import torch
 
 from fireredasr.data.asr_feat import ASRFeatExtractor
 from fireredasr.models.fireredasr_aed import FireRedAsrAed
-from fireredasr.models.fireredasr_llm import FireRedAsrLlm
 from fireredasr.tokenizer.aed_tokenizer import ChineseCharEnglishSpmTokenizer
-from fireredasr.tokenizer.llm_tokenizer import LlmTokenizerWrapper
 
 
 class FireRedAsr:
@@ -75,6 +73,8 @@ class FireRedAsr:
             return results
 
         elif self.asr_type == "llm":
+            from fireredasr.tokenizer.llm_tokenizer import LlmTokenizerWrapper
+
             input_ids, attention_mask, _, _ = \
                 LlmTokenizerWrapper.preprocess_texts(
                     origin_texts=[""]*feats.size(0), tokenizer=self.tokenizer,
@@ -115,6 +115,9 @@ def load_fireredasr_aed_model(model_path):
 
 
 def load_firered_llm_model_and_tokenizer(model_path, encoder_path, llm_dir):
+    from fireredasr.models.fireredasr_llm import FireRedAsrLlm
+    from fireredasr.tokenizer.llm_tokenizer import LlmTokenizerWrapper
+
     package = torch.load(model_path, map_location=lambda storage, loc: storage)
     package["args"].encoder_path = encoder_path
     package["args"].llm_dir = llm_dir

requirements.txt

@@ -1,11 +1,15 @@
-cn2an>=0.5.23
+# AX650N AED inference runtime.
 kaldiio>=2.18.0
 kaldi_native_fbank>=1.15
 numpy>=1.26.1
-peft>=0.13.2
 sentencepiece
-torch>=2.0.0
-transformers>=4.46.3
+setuptools<81
+torch==2.6.0
+torchaudio==2.6.0
 axengine @ https://github.com/AXERA-TECH/pyaxengine/releases/download/0.1.3.rc2/axengine-0.1.3-py3-none-any.whl
 silero_vad_axera
-torchaudio==2.6
+
+# Optional tools:
+# - fireredasr_onnx.py: onnxruntime
+# - fireredasr/speech2text.py LLM path: transformers, peft
+# - fireredasr/utils/wer.py --do_tn 1: cn2an