Ebranchformer #1951
Merged
Ebranchformer #1951
Conversation
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
- so ebranchformer feature extraction can be configured from Python - the GlobCmvn is not needed, at it is a module in the OnnxEncoder
KarelVesely84
force-pushed
the
ebranchformer
branch
from
81d6a00 to
41d378a
Compare
|
Thanks! Will review it today. |
csukuangfj
reviewed
Mar 4, 2025
sherpa-onnx/csrc/features.cc
Outdated
| @@ -48,7 +48,9 @@ std::string FeatureExtractorConfig::ToString() const { | |||
| os << "feature_dim=" << feature_dim << ", "; | |||
| os << "low_freq=" << low_freq << ", "; | |||
| os << "high_freq=" << high_freq << ", "; | |||
| os << "dither=" << dither << ")"; | |||
| os << "dither=" << dither << ", "; | |||
| os << "normalize_samples=" << normalize_samples << ", "; | |||
There was a problem hiding this comment.
Suggested change
| os << "normalize_samples=" << normalize_samples << ", "; | |
| os << "normalize_samples=" << (normalize_samples ? "True" : "False" )<< ", "; |
sherpa-onnx/csrc/features.cc
Outdated
| os << "dither=" << dither << ")"; | ||
| os << "dither=" << dither << ", "; | ||
| os << "normalize_samples=" << normalize_samples << ", "; | ||
| os << "snip_edges=" << snip_edges << ")"; |
There was a problem hiding this comment.
Suggested change
| os << "snip_edges=" << snip_edges << ")"; | |
| os << "snip_edges=" << (snip_edges ? "True" : "False") << ")"; |
| @@ -34,6 +51,9 @@ void PybindOnlineStream(py::module *m) { | |||
| py::arg("sample_rate"), py::arg("waveform"), kAcceptWaveformUsage, | |||
| py::call_guard<py::gil_scoped_release>()) | |||
| .def("input_finished", &PyClass::InputFinished, | |||
| py::call_guard<py::gil_scoped_release>()) | |||
| .def("get_frames", &PyClass::GetFrames, | |||
| py::arg("frame_index"), py::arg("n"), | |||
There was a problem hiding this comment.
Suggested change
| py::arg("frame_index"), py::arg("n"), | |
| py::arg("frame_index"), py::arg("n"), kGetFramesUsage |
so that if you use help(OnlineStream.get_frames), you can view the help info in python.
| @@ -92,6 +96,8 @@ std::unique_ptr<OnlineTransducerModel> OnlineTransducerModel::Create( | |||
| const auto &model_type = config.model_type; | |||
| if (model_type == "conformer") { | |||
| return std::make_unique<OnlineConformerTransducerModel>(config); | |||
| } else if (model_type == "ebranchformer") { | |||
There was a problem hiding this comment.
Can you also update
sherpa-onnx/sherpa-onnx/csrc/online-transducer-model.cc
Lines 180 to 181 in 41d378a
| @@ -115,6 +121,8 @@ std::unique_ptr<OnlineTransducerModel> OnlineTransducerModel::Create( | |||
| switch (model_type) { | |||
| case ModelType::kConformer: | |||
| return std::make_unique<OnlineConformerTransducerModel>(config); | |||
| case ModelType::kEbranchformer: | |||
There was a problem hiding this comment.
Can you also update
sherpa-onnx/sherpa-onnx/csrc/online-transducer-model.cc
Lines 199 to 200 in 41d378a
It is for Android and HarmonyOS.
|
(Please ignore the failed CI tests.) |
|
ok, good, thank you for the feedback, |
csukuangfj
approved these changes
Mar 4, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Hello Fangyun @csukuangfj ,
i did extend sherpa-onnx to support our EBranchformer encoder implementation
that we currently use widely at Brno University in Technology.
The EBranchformer code is based on Conformer
modelfromtransformers, but the internals are different:https://github.com/BUTSpeechFIT/huggingface_asr/blob/streaming_karel/src/models/encoders/e_branchformer.py
This allows to pre-train the encoder with BestRQ alg. and then fine-tune with modified icefall.
We would like to deploy it as a production system for streaming ASR (it already works for me locally).
So for us it would be good to have the support directly inside sherpa-onnx,
so we can use the official sherpa-onnx builds.
On the other hand, it is a bit specific model, not yet widespread.
Would you agree on accepting this extension into the codebase ?
Or, should we rely on our cusom builds ?
The Encoder assumes little different preset of input features derived from
Speech2TextFeatureExtractor, hence newly surfacing the FBANK options:
normalize_samplesandsnip_edges.Best regards,
Karel