Add runtime support for wespeaker models (#516)

sherpa-onnx/python/csrc/CMakeLists.txt

@@ -30,6 +30,8 @@ pybind11_add_module(_sherpa_onnx
   online-zipformer2-ctc-model-config.cc
   sherpa-onnx.cc
   silero-vad-model-config.cc
+  speaker-embedding-extractor.cc
+  speaker-embedding-manager.cc
   vad-model-config.cc
   vad-model.cc
   voice-activity-detector.cc

sherpa-onnx/python/csrc/online-recognizer.h

@@ -1,4 +1,4 @@
-// sherpa-onnx/python/csrc/online-recongizer.h
+// sherpa-onnx/python/csrc/online-recognizer.h
 //
 // Copyright (c)  2023  Xiaomi Corporation
 

sherpa-onnx/python/csrc/sherpa-onnx.cc

@@ -18,6 +18,8 @@
 #include "sherpa-onnx/python/csrc/online-model-config.h"
 #include "sherpa-onnx/python/csrc/online-recognizer.h"
 #include "sherpa-onnx/python/csrc/online-stream.h"
+#include "sherpa-onnx/python/csrc/speaker-embedding-extractor.h"
+#include "sherpa-onnx/python/csrc/speaker-embedding-manager.h"
 #include "sherpa-onnx/python/csrc/vad-model-config.h"
 #include "sherpa-onnx/python/csrc/vad-model.h"
 #include "sherpa-onnx/python/csrc/voice-activity-detector.h"
@@ -48,6 +50,8 @@ PYBIND11_MODULE(_sherpa_onnx, m) {
   PybindVoiceActivityDetector(&m);
 
   PybindOfflineTts(&m);
+  PybindSpeakerEmbeddingExtractor(&m);
+  PybindSpeakerEmbeddingManager(&m);
 }
 
 }  // namespace sherpa_onnx

sherpa-onnx/python/csrc/speaker-embedding-extractor.cc

@@ -0,0 +1,44 @@
+// sherpa-onnx/python/csrc/speaker-embedding-extractor.cc
+//
+// Copyright (c)  2023  Xiaomi Corporation
+
+#include "sherpa-onnx/python/csrc/speaker-embedding-extractor.h"
+
+#include <string>
+
+#include "sherpa-onnx/csrc/speaker-embedding-extractor.h"
+
+namespace sherpa_onnx {
+
+static void PybindSpeakerEmbeddingExtractorConfig(py::module *m) {
+  using PyClass = SpeakerEmbeddingExtractorConfig;
+  py::class_<PyClass>(*m, "SpeakerEmbeddingExtractorConfig")
+      .def(py::init<>())
+      .def(py::init<const std::string &, int32_t, bool, const std::string>(),
+           py::arg("model"), py::arg("num_threads") = 1,
+           py::arg("debug") = false, py::arg("provider") = "cpu")
+      .def_readwrite("model", &PyClass::model)
+      .def_readwrite("num_threads", &PyClass::num_threads)
+      .def_readwrite("debug", &PyClass::debug)
+      .def_readwrite("provider", &PyClass::provider)
+      .def("validate", &PyClass::Validate)
+      .def("__str__", &PyClass::ToString);
+}
+
+void PybindSpeakerEmbeddingExtractor(py::module *m) {
+  PybindSpeakerEmbeddingExtractorConfig(m);
+
+  using PyClass = SpeakerEmbeddingExtractor;
+  py::class_<PyClass>(*m, "SpeakerEmbeddingExtractor")
+      .def(py::init<const SpeakerEmbeddingExtractorConfig &>(),
+           py::arg("config"), py::call_guard<py::gil_scoped_release>())
+      .def_property_readonly("dim", &PyClass::Dim)
+      .def("create_stream", &PyClass::CreateStream,
+           py::call_guard<py::gil_scoped_release>())
+      .def("compute", &PyClass::Compute,
+           py::call_guard<py::gil_scoped_release>())
+      .def("is_ready", &PyClass::IsReady,
+           py::call_guard<py::gil_scoped_release>());
+}
+
+}  // namespace sherpa_onnx

sherpa-onnx/python/csrc/speaker-embedding-extractor.h

@@ -0,0 +1,16 @@
+// sherpa-onnx/python/csrc/speaker-embedding-extractor.h
+//
+// Copyright (c)  2023  Xiaomi Corporation
+
+#ifndef SHERPA_ONNX_PYTHON_CSRC_SPEAKER_EMBEDDING_EXTRACTOR_H_
+#define SHERPA_ONNX_PYTHON_CSRC_SPEAKER_EMBEDDING_EXTRACTOR_H_
+
+#include "sherpa-onnx/python/csrc/sherpa-onnx.h"
+
+namespace sherpa_onnx {
+
+void PybindSpeakerEmbeddingExtractor(py::module *m);
+
+}
+
+#endif  // SHERPA_ONNX_PYTHON_CSRC_SPEAKER_EMBEDDING_EXTRACTOR_H_

sherpa-onnx/python/csrc/speaker-embedding-manager.cc

@@ -0,0 +1,50 @@
+// sherpa-onnx/python/csrc/speaker-embedding-manager.cc
+//
+// Copyright (c)  2023  Xiaomi Corporation
+
+#include "sherpa-onnx/python/csrc/speaker-embedding-manager.h"
+
+#include <string>
+#include <vector>
+
+#include "sherpa-onnx/csrc/speaker-embedding-manager.h"
+
+namespace sherpa_onnx {
+
+void PybindSpeakerEmbeddingManager(py::module *m) {
+  using PyClass = SpeakerEmbeddingManager;
+  py::class_<PyClass>(*m, "SpeakerEmbeddingManager")
+      .def(py::init<int32_t>(), py::arg("dim"),
+           py::call_guard<py::gil_scoped_release>())
+      .def_property_readonly("num_speakers", &PyClass::NumSpeakers)
+      .def(
+          "add",
+          [](const PyClass &self, const std::string &name,
+             const std::vector<float> &v) -> bool {
+            return self.Add(name, v.data());
+          },
+          py::arg("name"), py::arg("v"),
+          py::call_guard<py::gil_scoped_release>())
+      .def(
+          "remove",
+          [](const PyClass &self, const std::string &name) -> bool {
+            return self.Remove(name);
+          },
+          py::arg("name"), py::call_guard<py::gil_scoped_release>())
+      .def(
+          "search",
+          [](const PyClass &self, const std::vector<float> &v, float threshold)
+              -> std::string { return self.Search(v.data(), threshold); },
+          py::arg("v"), py::arg("threshold"),
+          py::call_guard<py::gil_scoped_release>())
+      .def(
+          "verify",
+          [](const PyClass &self, const std::string &name,
+             const std::vector<float> &v, float threshold) -> bool {
+            return self.Verify(name, v.data(), threshold);
+          },
+          py::arg("name"), py::arg("v"), py::arg("threshold"),
+          py::call_guard<py::gil_scoped_release>());
+}
+
+}  // namespace sherpa_onnx

sherpa-onnx/python/csrc/speaker-embedding-manager.h

@@ -0,0 +1,16 @@
+// sherpa-onnx/python/csrc/speaker-embedding-manager.h
+//
+// Copyright (c)  2023  Xiaomi Corporation
+
+#ifndef SHERPA_ONNX_PYTHON_CSRC_SPEAKER_EMBEDDING_MANAGER_H_
+#define SHERPA_ONNX_PYTHON_CSRC_SPEAKER_EMBEDDING_MANAGER_H_
+
+#include "sherpa-onnx/python/csrc/sherpa-onnx.h"
+
+namespace sherpa_onnx {
+
+void PybindSpeakerEmbeddingManager(py::module *m);
+
+}  // namespace sherpa_onnx
+
+#endif  // SHERPA_ONNX_PYTHON_CSRC_SPEAKER_EMBEDDING_MANAGER_H_

sherpa-onnx/python/csrc/voice-activity-detector.cc

@@ -32,6 +32,7 @@ void PybindVoiceActivityDetector(py::module *m) {
             self.AcceptWaveform(samples.data(), samples.size());
           },
           py::arg("samples"), py::call_guard<py::gil_scoped_release>())
+      .def_property_readonly("config", &PyClass::GetConfig)
       .def("empty", &PyClass::Empty, py::call_guard<py::gil_scoped_release>())
       .def("pop", &PyClass::Pop, py::call_guard<py::gil_scoped_release>())
       .def("is_speech_detected", &PyClass::IsSpeechDetected,

sherpa-onnx/python/sherpa_onnx/__init__.py

@@ -8,6 +8,9 @@ from _sherpa_onnx import (
     OfflineTtsVitsModelConfig,
     OnlineStream,
     SileroVadModelConfig,
+    SpeakerEmbeddingExtractor,
+    SpeakerEmbeddingExtractorConfig,
+    SpeakerEmbeddingManager,
     SpeechSegment,
     VadModel,
     VadModelConfig,