-
-
Notifications
You must be signed in to change notification settings - Fork 183
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add sherpa-onnx support #50
Merged
+711
−1
Merged
Changes from all commits
Commits
File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,155 @@ | ||
| import numpy as np | ||
| import sherpa_onnx | ||
| from .asr_interface import ASRInterface | ||
|
|
||
| class VoiceRecognition(ASRInterface): | ||
|
|
||
| def __init__( | ||
| self, | ||
| model_type: str = "paraformer", # or "transducer", "nemo_ctc", "wenet_ctc", "whisper", "tdnn_ctc", "sense_voice" | ||
| encoder: str = None, # Path to the encoder model, used with transducer | ||
| decoder: str = None, # Path to the decoder model, used with transducer | ||
| joiner: str = None, # Path to the joiner model, used with transducer | ||
| paraformer: str = None, # Path to the model.onnx from Paraformer | ||
| nemo_ctc: str = None, # Path to the model.onnx from NeMo CTC | ||
| wenet_ctc: str = None, # Path to the model.onnx from WeNet CTC | ||
| tdnn_model: str = None, # Path to the model.onnx for the tdnn model of the yesno recipe | ||
| whisper_encoder: str = None, # Path to whisper encoder model | ||
| whisper_decoder: str = None, # Path to whisper decoder model | ||
| sense_voice: str = None, # Path to the model.onnx from SenseVoice | ||
| tokens: str = None, # Path to tokens.txt | ||
| hotwords_file: str = "", # Path to hotwords file | ||
| hotwords_score: float = 1.5, # Hotwords score | ||
| modeling_unit: str = "", # Modeling unit for hotwords | ||
| bpe_vocab: str = "", # Path to bpe vocabulary, used with hotwords | ||
| num_threads: int = 1, # Number of threads for neural network computation | ||
| whisper_language: str = "", # Language for whisper model | ||
| whisper_task: str = "transcribe", # Task for whisper model (transcribe or translate) | ||
| whisper_tail_paddings: int = -1, # Tail padding frames for whisper model | ||
| blank_penalty: float = 0.0, # Penalty for blank symbol | ||
| decoding_method: str = "greedy_search", # Decoding method (greedy_search or modified_beam_search) | ||
| debug: bool = False, # Show debug messages | ||
| sample_rate: int = 16000, # Sample rate | ||
| feature_dim: int = 80, # Feature dimension | ||
| use_itn: bool = True, # Use ITN for SenseVoice models | ||
| ) -> None: | ||
|
|
||
| self.model_type = model_type | ||
| self.encoder = encoder | ||
| self.decoder = decoder | ||
| self.joiner = joiner | ||
| self.paraformer = paraformer | ||
| self.nemo_ctc = nemo_ctc | ||
| self.wenet_ctc = wenet_ctc | ||
| self.tdnn_model = tdnn_model | ||
| self.whisper_encoder = whisper_encoder | ||
| self.whisper_decoder = whisper_decoder | ||
| self.sense_voice = sense_voice | ||
| self.tokens = tokens | ||
| self.hotwords_file = hotwords_file | ||
| self.hotwords_score = hotwords_score | ||
| self.modeling_unit = modeling_unit | ||
| self.bpe_vocab = bpe_vocab | ||
| self.num_threads = num_threads | ||
| self.whisper_language = whisper_language | ||
| self.whisper_task = whisper_task | ||
| self.whisper_tail_paddings = whisper_tail_paddings | ||
| self.blank_penalty = blank_penalty | ||
| self.decoding_method = decoding_method | ||
| self.debug = debug | ||
| self.SAMPLE_RATE = sample_rate | ||
| self.feature_dim = feature_dim | ||
| self.use_itn = use_itn | ||
|
|
||
| self.asr_with_vad = None | ||
|
|
||
| self.recognizer = self._create_recognizer() | ||
|
|
||
| def _create_recognizer(self): | ||
| if self.model_type == "transducer": | ||
| recognizer = sherpa_onnx.OfflineRecognizer.from_transducer( | ||
| encoder=self.encoder, | ||
| decoder=self.decoder, | ||
| joiner=self.joiner, | ||
| tokens=self.tokens, | ||
| num_threads=self.num_threads, | ||
| sample_rate=self.SAMPLE_RATE, | ||
| feature_dim=self.feature_dim, | ||
| decoding_method=self.decoding_method, | ||
| hotwords_file=self.hotwords_file, | ||
| hotwords_score=self.hotwords_score, | ||
| modeling_unit=self.modeling_unit, | ||
| bpe_vocab=self.bpe_vocab, | ||
| blank_penalty=self.blank_penalty, | ||
| debug=self.debug, | ||
| ) | ||
| elif self.model_type == "paraformer": | ||
| recognizer = sherpa_onnx.OfflineRecognizer.from_paraformer( | ||
| paraformer=self.paraformer, | ||
| tokens=self.tokens, | ||
| num_threads=self.num_threads, | ||
| sample_rate=self.SAMPLE_RATE, | ||
| feature_dim=self.feature_dim, | ||
| decoding_method=self.decoding_method, | ||
| debug=self.debug, | ||
| ) | ||
| elif self.model_type == "nemo_ctc": | ||
| recognizer = sherpa_onnx.OfflineRecognizer.from_nemo_ctc( | ||
| model=self.nemo_ctc, | ||
| tokens=self.tokens, | ||
| num_threads=self.num_threads, | ||
| sample_rate=self.SAMPLE_RATE, | ||
| feature_dim=self.feature_dim, | ||
| decoding_method=self.decoding_method, | ||
| debug=self.debug, | ||
| ) | ||
| elif self.model_type == "wenet_ctc": | ||
| recognizer = sherpa_onnx.OfflineRecognizer.from_wenet_ctc( | ||
| model=self.wenet_ctc, | ||
| tokens=self.tokens, | ||
| num_threads=self.num_threads, | ||
| sample_rate=self.SAMPLE_RATE, | ||
| feature_dim=self.feature_dim, | ||
| decoding_method=self.decoding_method, | ||
| debug=self.debug, | ||
| ) | ||
| elif self.model_type == "whisper": | ||
| recognizer = sherpa_onnx.OfflineRecognizer.from_whisper( | ||
| encoder=self.whisper_encoder, | ||
| decoder=self.whisper_decoder, | ||
| tokens=self.tokens, | ||
| num_threads=self.num_threads, | ||
| decoding_method=self.decoding_method, | ||
| debug=self.debug, | ||
| language=self.whisper_language, | ||
| task=self.whisper_task, | ||
| tail_paddings=self.whisper_tail_paddings, | ||
| ) | ||
| elif self.model_type == "tdnn_ctc": | ||
| recognizer = sherpa_onnx.OfflineRecognizer.from_tdnn_ctc( | ||
| model=self.tdnn_model, | ||
| tokens=self.tokens, | ||
| sample_rate=self.SAMPLE_RATE, | ||
| feature_dim=self.feature_dim, | ||
| num_threads=self.num_threads, | ||
| decoding_method=self.decoding_method, | ||
| debug=self.debug, | ||
| ) | ||
| elif self.model_type == "sense_voice": | ||
| recognizer = sherpa_onnx.OfflineRecognizer.from_sense_voice( | ||
| model=self.sense_voice, | ||
| tokens=self.tokens, | ||
| num_threads=self.num_threads, | ||
| use_itn=self.use_itn, | ||
| debug=self.debug | ||
| ) | ||
| else: | ||
| raise ValueError(f"Invalid model type: {self.model_type}") | ||
|
|
||
| return recognizer | ||
|
|
||
| def transcribe_np(self, audio: np.ndarray) -> str: | ||
| stream = self.recognizer.create_stream() | ||
| stream.accept_waveform(self.SAMPLE_RATE, audio) | ||
| self.recognizer.decode_streams([stream]) | ||
| return stream.result.text |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I suggest that you use model.int8.onnx, which is way smaller in file size than that of model.onnx.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the heads up! I appreciate the suggestion to use model.int8.onnx. Do you have any insights into how the int8 model performs compared to the original model.onnx in terms of recognition accuracy? Have you had a chance to evaluate its performance?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I only tested it on fewer than 4 test wave files and the quantized model produces identical results as the not quantized model.