Add Python ASR example with alsa (#324)

CMakeLists.txt

@@ -1,7 +1,7 @@
 cmake_minimum_required(VERSION 3.13 FATAL_ERROR)
 project(sherpa-ncnn)
 
-set(SHERPA_NCNN_VERSION "2.1.9")
+set(SHERPA_NCNN_VERSION "2.1.10")
 
 # Disable warning about
 #
@@ -106,6 +106,7 @@ if(SHERPA_NCNN_ENABLE_BINARY AND UNIX AND NOT APPLE)
   include(CheckIncludeFileCXX)
   check_include_file_cxx(alsa/asoundlib.h SHERPA_NCNN_HAS_ALSA)
   if(SHERPA_NCNN_HAS_ALSA)
+    message(STATUS "With Alsa")
     add_definitions(-DSHERPA_NCNN_ENABLE_ALSA=1)
   elseif(UNIX AND NOT APPLE)
     message(WARNING "\

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

@@ -0,0 +1,111 @@
+#!/usr/bin/env python3
+
+# Real-time speech recognition from a microphone with sherpa-ncnn 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/ncnn/pretrained_models/index.html
+# to download pre-trained models
+
+import argparse
+import sys
+
+import sherpa_ncnn
+
+
+def get_args():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    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():
+    # Please replace the model files if needed.
+    # See https://k2-fsa.github.io/sherpa/ncnn/pretrained_models/index.html
+    # for download links.
+    recognizer = sherpa_ncnn.Recognizer(
+        tokens="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/tokens.txt",
+        encoder_param="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/encoder_jit_trace-pnnx.ncnn.param",
+        encoder_bin="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/encoder_jit_trace-pnnx.ncnn.bin",
+        decoder_param="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/decoder_jit_trace-pnnx.ncnn.param",
+        decoder_bin="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/decoder_jit_trace-pnnx.ncnn.bin",
+        joiner_param="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/joiner_jit_trace-pnnx.ncnn.param",
+        joiner_bin="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/joiner_jit_trace-pnnx.ncnn.bin",
+        num_threads=4,
+        decoding_method="modified_beam_search",
+        enable_endpoint_detection=True,
+        rule1_min_trailing_silence=2.4,
+        rule2_min_trailing_silence=1.2,
+        rule3_min_utterance_length=300,
+        hotwords_file="",
+        hotwords_score=1.5,
+    )
+    return recognizer
+
+
+def main():
+    args = get_args()
+    device_name = args.device_name
+    print(f"device_name: {device_name}")
+    alsa = sherpa_ncnn.Alsa(device_name)
+
+    recognizer = create_recognizer()
+    print("Started! Please speak")
+    sample_rate = recognizer.sample_rate
+    samples_per_read = int(0.1 * sample_rate)  # 0.1 second = 100 ms
+    last_result = ""
+    segment_id = 0
+
+    while True:
+        samples = alsa.read(samples_per_read)  # a blocking read
+        recognizer.accept_waveform(sample_rate, samples)
+
+        is_endpoint = recognizer.is_endpoint
+
+        result = recognizer.text
+        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()
+
+
+if __name__ == "__main__":
+    try:
+        main()
+    except KeyboardInterrupt:
+        print("\nCaught Ctrl + C. Exiting")

sherpa-ncnn/csrc/alsa.cc

@@ -62,7 +62,7 @@ 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
+  plughw:3,0
 
   )";
 

sherpa-ncnn/csrc/sherpa-ncnn-alsa.cc

@@ -67,7 +67,7 @@ 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
+  plughw:3,0
 
 as the device_name.
 )usage";

sherpa-ncnn/csrc/stream.cc

@@ -19,6 +19,7 @@
 #include "sherpa-ncnn/csrc/stream.h"
 
 #include <iostream>
+#include <utility>
 
 namespace sherpa_ncnn {
 

sherpa-ncnn/python/csrc/CMakeLists.txt

@@ -11,6 +11,12 @@ set(srcs
   stream.cc
 )
 
+if(SHERPA_NCNN_HAS_ALSA)
+  list(APPEND srcs ${CMAKE_SOURCE_DIR}/sherpa-ncnn/csrc/alsa.cc alsa.cc)
+else()
+  list(APPEND srcs faked-alsa.cc)
+endif()
+
 pybind11_add_module(_sherpa_ncnn ${srcs})
 target_link_libraries(_sherpa_ncnn PRIVATE sherpa-ncnn-core)
 
@@ -28,6 +34,13 @@ if(NOT WIN32)
   target_link_libraries(_sherpa_ncnn PRIVATE "-Wl,-rpath,${SHERPA_NCNN_RPATH_ORIGIN}/sherpa_ncnn/lib")
 endif()
 
+if(SHERPA_NCNN_HAS_ALSA)
+  if(DEFINED ENV{SHERPA_NCNN_ALSA_LIB_DIR})
+    target_link_libraries(_sherpa_ncnn PRIVATE -L$ENV{SHERPA_NCNN_ALSA_LIB_DIR} -lasound)
+  else()
+    target_link_libraries(_sherpa_ncnn PRIVATE asound)
+  endif()
+endif()
 
 install(TARGETS _sherpa_ncnn
   DESTINATION ../

sherpa-ncnn/python/csrc/alsa.cc

@@ -0,0 +1,30 @@
+// sherpa-ncnn/python/csrc/alsa.cc
+//
+// Copyright (c)  2024  Xiaomi Corporation
+
+#include "sherpa-ncnn/python/csrc/alsa.h"
+
+#include <vector>
+
+#include "sherpa-ncnn/csrc/alsa.h"
+
+namespace sherpa_ncnn {
+
+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_ncnn

sherpa-ncnn/python/csrc/alsa.h

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

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

@@ -0,0 +1,47 @@
+// sherpa-ncnn/python/csrc/faked-alsa.cc
+//
+// Copyright (c)  2024  Xiaomi Corporation
+
+#include "sherpa-ncnn/csrc/macros.h"
+#include "sherpa-ncnn/python/csrc/alsa.h"
+
+namespace sherpa_ncnn {
+
+class FakedAlsa {
+ public:
+  explicit FakedAlsa(const char *) {
+    SHERPA_NCNN_LOGE("This function is for Linux only.");
+#if (SHERPA_NCNN_ENABLE_ALSA == 0) && (defined(__unix__) || defined(__unix))
+    SHERPA_NCNN_LOGE(R"doc(
+sherpa-ncnn is compiled without alsa support. To enable that, please run
+  (1) sudo apt-get install alsa-utils libasound2-dev
+  (2) rebuild sherpa-ncnn
+)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_ncnn
+
+#endif  // SHERPA_NCNN_PYTHON_CSRC_FAKED_ALSA_H_

sherpa-ncnn/python/csrc/sherpa-ncnn.cc

@@ -18,6 +18,7 @@
 
 #include "sherpa-ncnn/python/csrc/sherpa-ncnn.h"
 
+#include "sherpa-ncnn/python/csrc/alsa.h"
 #include "sherpa-ncnn/python/csrc/decoder.h"
 #include "sherpa-ncnn/python/csrc/display.h"
 #include "sherpa-ncnn/python/csrc/endpoint.h"
@@ -39,6 +40,8 @@ PYBIND11_MODULE(_sherpa_ncnn, m) {
   PybindRecognizer(&m);
 
   PybindDisplay(&m);
+
+  PybindAlsa(&m);
 }
 
 }  // namespace sherpa_ncnn

sherpa-ncnn/python/sherpa_ncnn/__init__.py

@@ -1,2 +1,3 @@
+from _sherpa_ncnn import Alsa, Display
+
 from .recognizer import Recognizer
-from _sherpa_ncnn import Display