RNNT EOU 成功推理

examples/asr/asr_eou/speech_to_text_rnnt_eou_train.py

@@ -360,7 +360,7 @@ if __name__ == '__main__':
         'exp_manager.resume_ignore_no_checkpoint=true',
         'exp_manager.exp_dir=results/',
         'exp_manager.checkpoint_callback_params.save_top_k=1',
-        '++trainer.check_val_every_n_epoch=350',
+        '++trainer.check_val_every_n_epoch=1',
         '++model.encoder.conv_norm_type=layer_norm',
         'model.tokenizer.dir=data/common_voice_11_0/ja/tokenizers/tokenizer_spe_bpe_v2491_eou',
         'model.train_ds.tarred_audio_filepaths=data/common_voice_11_0/ja/train_tarred_1bk/audio__OP_0..1023_CL_.tar',
@@ -371,14 +371,14 @@ if __name__ == '__main__':
         '~model.train_ds.augmentor.noise',
         'model.validation_ds.manifest_filepath=data/common_voice_11_0/ja/validation_tarred_1bk/tarred_audio_manifest.json',
         'model.test_ds.manifest_filepath=data/common_voice_11_0/ja/test_tarred_1bk/tarred_audio_manifest.json',
-        'trainer.max_epochs=350',
+        'trainer.max_epochs=1',
         'trainer.devices=1',
         'trainer.accelerator=gpu',
         'trainer.strategy=auto',
         'trainer.log_every_n_steps=1',
-        'model.train_ds.batch_size=4', # Reduced batch size for transducer/EOU as it consumes more memory
-        'model.validation_ds.batch_size=4',
-        'model.test_ds.batch_size=4',
+        'model.train_ds.batch_size=1', # Reduced batch size for transducer/EOU as it consumes more memory
+        'model.validation_ds.batch_size=1',
+        'model.test_ds.batch_size=1',
         'model.train_ds.num_workers=0',
         'model.validation_ds.num_workers=0',
         'model.test_ds.num_workers=0',

readme.txt

@@ -291,7 +291,8 @@ TODO:
 
 
 
-
+训练 RNNT EOU
+python examples/asr/asr_eou/speech_to_text_rnnt_eou_train.py
 
 
 

server_aware_streaming.py

@@ -28,8 +28,8 @@ def _hash_audio(audio: np.ndarray) -> str:
 
 
 # DEFAULT_MODEL = "./nemotron-speech-streaming-en-0.6b/nemotron-speech-streaming-en-0.6b.nemo"
-DEFAULT_MODEL = "results/NeMo_Ja_FastConformer_Streaming/checkpoints/NeMo_Ja_FastConformer_Streaming.nemo"
-
+# DEFAULT_MODEL = "results/NeMo_Ja_FastConformer_Streaming/checkpoints/NeMo_Ja_FastConformer_Streaming.nemo"
+DEFAULT_MODEL = "results/NeMo_Ja_FastConformer_Transducer_RNNT_EOU/checkpoints/NeMo_Ja_FastConformer_Transducer_RNNT_EOU.nemo"
 
 # Right context options for att_context_size=[70, X]
 RIGHT_CONTEXT_OPTIONS = {

test_websocket_client.py

@@ -174,7 +174,7 @@ if __name__ == "__main__":
     if "--all" in sys.argv:
         asyncio.run(test_multiple_chunk_sizes(audio_path, server_url))
     else:
-        chunk_ms = 700
+        chunk_ms = 200
         if "--chunk" in sys.argv:
             idx = sys.argv.index("--chunk")
             if idx + 1 < len(sys.argv):