Add Python ASR examples with alsa (#646)

CMakeLists.txt

@@ -146,6 +146,7 @@ include(CheckIncludeFileCXX)
 if(UNIX AND NOT APPLE AND NOT SHERPA_ONNX_ENABLE_WASM AND NOT CMAKE_SYSTEM_NAME STREQUAL Android)
   check_include_file_cxx(alsa/asoundlib.h SHERPA_ONNX_HAS_ALSA)
   if(SHERPA_ONNX_HAS_ALSA)
+    message(STATUS "With Alsa")
     add_definitions(-DSHERPA_ONNX_ENABLE_ALSA=1)
   else()
     message(WARNING "\

python-api-examples/speech-recognition-from-microphone-with-endpoint-detection-alsa.py

@@ -0,0 +1,207 @@
+#!/usr/bin/env python3
+
+# Real-time speech recognition from a microphone with sherpa-onnx Python API
+# with endpoint detection.
+#
+# Note: This script uses ALSA and works only on Linux systems, especially
+# for embedding Linux systems and for running Linux on Windows using WSL.
+#
+# Please refer to
+# https://k2-fsa.github.io/sherpa/onnx/pretrained_models/index.html
+# to download pre-trained models
+
+import argparse
+import sys
+from pathlib import Path
+import sherpa_onnx
+
+
+def assert_file_exists(filename: str):
+    assert Path(filename).is_file(), (
+        f"{filename} does not exist!\n"
+        "Please refer to "
+        "https://k2-fsa.github.io/sherpa/onnx/pretrained_models/index.html to download it"
+    )
+
+
+def get_args():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    parser.add_argument(
+        "--tokens",
+        type=str,
+        required=True,
+        help="Path to tokens.txt",
+    )
+
+    parser.add_argument(
+        "--encoder",
+        type=str,
+        required=True,
+        help="Path to the encoder model",
+    )
+
+    parser.add_argument(
+        "--decoder",
+        type=str,
+        required=True,
+        help="Path to the decoder model",
+    )
+
+    parser.add_argument(
+        "--joiner",
+        type=str,
+        required=True,
+        help="Path to the joiner model",
+    )
+
+    parser.add_argument(
+        "--decoding-method",
+        type=str,
+        default="greedy_search",
+        help="Valid values are greedy_search and modified_beam_search",
+    )
+
+    parser.add_argument(
+        "--provider",
+        type=str,
+        default="cpu",
+        help="Valid values: cpu, cuda, coreml",
+    )
+
+    parser.add_argument(
+        "--hotwords-file",
+        type=str,
+        default="",
+        help="""
+        The file containing hotwords, one words/phrases per line, and for each
+        phrase the bpe/cjkchar are separated by a space. For example:
+
+        ▁HE LL O ▁WORLD
+        你 好 世 界
+        """,
+    )
+
+    parser.add_argument(
+        "--hotwords-score",
+        type=float,
+        default=1.5,
+        help="""
+        The hotword score of each token for biasing word/phrase. Used only if
+        --hotwords-file is given.
+        """,
+    )
+
+    parser.add_argument(
+        "--blank-penalty",
+        type=float,
+        default=0.0,
+        help="""
+        The penalty applied on blank symbol during decoding.
+        Note: It is a positive value that would be applied to logits like
+        this `logits[:, 0] -= blank_penalty` (suppose logits.shape is
+        [batch_size, vocab] and blank id is 0).
+        """,
+    )
+
+    parser.add_argument(
+        "--device-name",
+        type=str,
+        required=True,
+        help="""
+The device name specifies which microphone to use in case there are several
+on your system. You can use
+
+  arecord -l
+
+to find all available microphones on your computer. For instance, if it outputs
+
+**** List of CAPTURE Hardware Devices ****
+card 3: UACDemoV10 [UACDemoV1.0], device 0: USB Audio [USB Audio]
+  Subdevices: 1/1
+  Subdevice #0: subdevice #0
+
+and if you want to select card 3 and the device 0 on that card, please use:
+
+  plughw:3,0
+
+as the device_name.
+        """,
+    )
+
+    return parser.parse_args()
+
+
+def create_recognizer(args):
+    assert_file_exists(args.encoder)
+    assert_file_exists(args.decoder)
+    assert_file_exists(args.joiner)
+    assert_file_exists(args.tokens)
+    # Please replace the model files if needed.
+    # See https://k2-fsa.github.io/sherpa/onnx/pretrained_models/index.html
+    # for download links.
+    recognizer = sherpa_onnx.OnlineRecognizer.from_transducer(
+        tokens=args.tokens,
+        encoder=args.encoder,
+        decoder=args.decoder,
+        joiner=args.joiner,
+        num_threads=1,
+        sample_rate=16000,
+        feature_dim=80,
+        enable_endpoint_detection=True,
+        rule1_min_trailing_silence=2.4,
+        rule2_min_trailing_silence=1.2,
+        rule3_min_utterance_length=300,  # it essentially disables this rule
+        decoding_method=args.decoding_method,
+        provider=args.provider,
+        hotwords_file=args.hotwords_file,
+        hotwords_score=args.hotwords_score,
+        blank_penalty=args.blank_penalty,
+    )
+    return recognizer
+
+
+def main():
+    args = get_args()
+    device_name = args.device_name
+    print(f"device_name: {device_name}")
+    alsa = sherpa_onnx.Alsa(device_name)
+
+    print("Creating recognizer")
+    recognizer = create_recognizer(args)
+    print("Started! Please speak")
+
+    sample_rate = 16000
+    samples_per_read = int(0.1 * sample_rate)  # 0.1 second = 100 ms
+
+    stream = recognizer.create_stream()
+
+    last_result = ""
+    segment_id = 0
+    while True:
+        samples = alsa.read(samples_per_read)  # a blocking read
+        stream.accept_waveform(sample_rate, samples)
+        while recognizer.is_ready(stream):
+            recognizer.decode_stream(stream)
+
+        is_endpoint = recognizer.is_endpoint(stream)
+
+        result = recognizer.get_result(stream)
+
+        if result and (last_result != result):
+            last_result = result
+            print("\r{}:{}".format(segment_id, result), end="", flush=True)
+        if is_endpoint:
+            if result:
+                print("\r{}:{}".format(segment_id, result), flush=True)
+                segment_id += 1
+            recognizer.reset(stream)
+
+
+if __name__ == "__main__":
+    try:
+        main()
+    except KeyboardInterrupt:
+        print("\nCaught Ctrl + C. Exiting")

sherpa-onnx/csrc/session.cc

@@ -16,7 +16,7 @@
 #endif
 
 #if __ANDROID_API__ >= 27
-#include "nnapi_provider_factory.h"
+#include "nnapi_provider_factory.h"  // NOLINT
 #endif
 
 namespace sherpa_onnx {

sherpa-onnx/csrc/sherpa-onnx-alsa-offline-speaker-identification.cc

@@ -276,8 +276,8 @@ as the device_name.
       }
     }
 
-    using namespace std::chrono_literals;
-    std::this_thread::sleep_for(20ms);  // sleep for 20ms
+    using namespace std::chrono_literals;  // NOLINT
+    std::this_thread::sleep_for(20ms);     // sleep for 20ms
   }
 
   t.join();

sherpa-onnx/csrc/sherpa-onnx-alsa-offline.cc

@@ -192,8 +192,8 @@ as the device_name.
       }
     }
 
-    using namespace std::chrono_literals;
-    std::this_thread::sleep_for(20ms);  // sleep for 20ms
+    using namespace std::chrono_literals;  // NOLINT
+    std::this_thread::sleep_for(20ms);     // sleep for 20ms
   }
   t.join();
   t2.join();

sherpa-onnx/csrc/sherpa-onnx-keyword-spotter-alsa.cc

@@ -53,10 +53,6 @@ card 3: UACDemoV10 [UACDemoV1.0], device 0: USB Audio [USB Audio]
 
 and if you want to select card 3 and the device 0 on that card, please use:
 
-  hw:3,0
-
-or
-
   plughw:3,0
 
 as the device_name.

sherpa-onnx/python/csrc/CMakeLists.txt

@@ -1,6 +1,6 @@
 include_directories(${CMAKE_SOURCE_DIR})
 
-pybind11_add_module(_sherpa_onnx
+set(srcs
   circular-buffer.cc
   display.cc
   endpoint.cc
@@ -37,6 +37,13 @@ pybind11_add_module(_sherpa_onnx
   vad-model.cc
   voice-activity-detector.cc
 )
+if(SHERPA_ONNX_HAS_ALSA)
+  list(APPEND srcs ${CMAKE_SOURCE_DIR}/sherpa-onnx/csrc/alsa.cc alsa.cc)
+else()
+  list(APPEND srcs faked-alsa.cc)
+endif()
+
+pybind11_add_module(_sherpa_onnx ${srcs})
 
 if(APPLE)
   execute_process(
@@ -54,6 +61,14 @@ endif()
 
 target_link_libraries(_sherpa_onnx PRIVATE sherpa-onnx-core)
 
+if(SHERPA_ONNX_HAS_ALSA)
+  if(DEFINED ENV{SHERPA_ONNX_ALSA_LIB_DIR})
+    target_link_libraries(_sherpa_onnx PRIVATE -L$ENV{SHERPA_ONNX_ALSA_LIB_DIR} -lasound)
+  else()
+    target_link_libraries(_sherpa_onnx PRIVATE asound)
+  endif()
+endif()
+
 install(TARGETS _sherpa_onnx
   DESTINATION ../
 )

sherpa-onnx/python/csrc/alsa.cc

@@ -0,0 +1,30 @@
+// sherpa-onnx/python/csrc/alsa.cc
+//
+// Copyright (c)  2024  Xiaomi Corporation
+
+#include "sherpa-onnx/python/csrc/alsa.h"
+
+#include <vector>
+
+#include "sherpa-onnx/csrc/alsa.h"
+
+namespace sherpa_onnx {
+
+void PybindAlsa(py::module *m) {
+  using PyClass = Alsa;
+  py::class_<PyClass>(*m, "Alsa")
+      .def(py::init<const char *>(), py::arg("device_name"),
+           py::call_guard<py::gil_scoped_release>())
+      .def(
+          "read",
+          [](PyClass &self, int32_t num_samples) -> std::vector<float> {
+            return self.Read(num_samples);
+          },
+          py::arg("num_samples"), py::call_guard<py::gil_scoped_release>())
+      .def_property_readonly("expected_sample_rate",
+                             &PyClass::GetExpectedSampleRate)
+      .def_property_readonly("actual_sample_rate",
+                             &PyClass::GetActualSampleRate);
+}
+
+}  // namespace sherpa_onnx

sherpa-onnx/python/csrc/alsa.h

@@ -0,0 +1,16 @@
+// sherpa-onnx/python/csrc/alsa.h
+//
+// Copyright (c)  2024  Xiaomi Corporation
+
+#ifndef SHERPA_ONNX_PYTHON_CSRC_ALSA_H_
+#define SHERPA_ONNX_PYTHON_CSRC_ALSA_H_
+
+#include "sherpa-onnx/python/csrc/sherpa-onnx.h"
+
+namespace sherpa_onnx {
+
+void PybindAlsa(py::module *m);
+
+}  // namespace sherpa_onnx
+
+#endif  // SHERPA_ONNX_PYTHON_CSRC_ALSA_H_

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

@@ -0,0 +1,47 @@
+// sherpa-onnx/python/csrc/faked-alsa.cc
+//
+// Copyright (c)  2024  Xiaomi Corporation
+
+#include "sherpa-onnx/csrc/macros.h"
+#include "sherpa-onnx/python/csrc/alsa.h"
+
+namespace sherpa_onnx {
+
+class FakedAlsa {
+ public:
+  explicit FakedAlsa(const char *) {
+    SHERPA_ONNX_LOGE("This function is for Linux only.");
+#if (SHERPA_ONNX_ENABLE_ALSA == 0) && (defined(__unix__) || defined(__unix))
+    SHERPA_ONNX_LOGE(R"doc(
+sherpa-onnx is compiled without alsa support. To enable that, please run
+  (1) sudo apt-get install alsa-utils libasound2-dev
+  (2) rebuild sherpa-onnx
+)doc");
+#endif
+    exit(-1);
+  }
+
+  std::vector<float> Read(int32_t) const { return {}; }
+  int32_t GetExpectedSampleRate() const { return -1; }
+  int32_t GetActualSampleRate() const { return -1; }
+};
+
+void PybindAlsa(py::module *m) {
+  using PyClass = FakedAlsa;
+  py::class_<PyClass>(*m, "Alsa")
+      .def(py::init<const char *>(), py::arg("device_name"))
+      .def(
+          "read",
+          [](PyClass &self, int32_t num_samples) -> std::vector<float> {
+            return self.Read(num_samples);
+          },
+          py::arg("num_samples"), py::call_guard<py::gil_scoped_release>())
+      .def_property_readonly("expected_sample_rate",
+                             &PyClass::GetExpectedSampleRate)
+      .def_property_readonly("actual_sample_rate",
+                             &PyClass::GetActualSampleRate);
+}
+
+}  // namespace sherpa_onnx
+
+#endif  // SHERPA_ONNX_PYTHON_CSRC_FAKED_ALSA_H_

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

@@ -4,6 +4,7 @@
 
 #include "sherpa-onnx/python/csrc/sherpa-onnx.h"
 
+#include "sherpa-onnx/python/csrc/alsa.h"
 #include "sherpa-onnx/python/csrc/circular-buffer.h"
 #include "sherpa-onnx/python/csrc/display.h"
 #include "sherpa-onnx/python/csrc/endpoint.h"
@@ -54,6 +55,8 @@ PYBIND11_MODULE(_sherpa_onnx, m) {
   PybindOfflineTts(&m);
   PybindSpeakerEmbeddingExtractor(&m);
   PybindSpeakerEmbeddingManager(&m);
+
+  PybindAlsa(&m);
 }
 
 }  // namespace sherpa_onnx

sherpa-onnx/python/sherpa_onnx/__init__.py

@@ -1,4 +1,5 @@
 from _sherpa_onnx import (
+    Alsa,
     CircularBuffer,
     Display,
     OfflineStream,