Merge branch 'feat/stream-audio' into 'dev'

Implement audio streaming

See merge request ics/sp/2025/n25b/pepperplus-ri!8

README.md

@@ -34,6 +34,18 @@ python -m virtualenv .venv
 source .venv/bin/activate
 ```
 
+We depend on PortAudio for the `pyaudio` package, so install it with:
+
+```bash
+sudo apt install -y portaudio19-dev
+```
+
+On WSL, also install:
+
+```bash
+sudo apt install -y libasound2-plugins
+```
+
 Install the required packages with
 
 ```bash
@@ -98,6 +110,8 @@ $env:PYTHONPATH="src"; python -m robot_interface.main
 
 With both, if you want to connect to the actual robot (or simulator), pass the `--qi-url` argument.
 
+There's also a `--microphone` argument that can be used to choose a microphone to use. If not given, the program will try the default microphone. If you don't know the name of the microphone, pass the argument with any value, and it will list the names of available microphones.
+
 
 
 ## Testing

src/robot_interface/endpoints/actuation_receiver.py

@@ -1,3 +1,4 @@
+from __future__ import unicode_literals  # So that we can log texts with Unicode characters
 import logging
 
 import zmq

src/robot_interface/endpoints/audio_sender.py

@@ -0,0 +1,72 @@
+from __future__ import unicode_literals  # So that `logging` can use Unicode characters in names
+import threading
+import logging
+
+import pyaudio
+import zmq
+
+from robot_interface.endpoints.socket_base import SocketBase
+from robot_interface.state import state
+from robot_interface.utils.microphone import choose_mic
+
+
+logger = logging.getLogger(__name__)
+
+
+class AudioSender(SocketBase):
+    def __init__(self, zmq_context, port=5558):
+        super(AudioSender, self).__init__(str("audio"))  # Convert future's unicode_literal to str
+        self.create_socket(zmq_context, zmq.PUB, port)
+        self.thread = None
+
+        try:
+            self.audio = pyaudio.PyAudio()
+            self.microphone = choose_mic(self.audio)
+        except IOError as e:
+            logger.warning("PyAudio is not available.", exc_info=e)
+            self.audio = None
+            self.microphone = None
+
+    def start(self):
+        """
+        Start sending audio in a different thread.
+        """
+        if not self.microphone:
+            logger.info("Not listening: no microphone available.")
+            return
+
+        logger.info("Listening with microphone \"{}\".".format(self.microphone["name"]))
+        self.thread = threading.Thread(target=self._stream)
+        self.thread.start()
+
+    def wait_until_done(self):
+        """
+        Wait until the audio thread is done. Will only be done if `state.exit_event` is set, so
+        make sure to set that before calling this method or it will block.
+        """
+        if not self.thread: return
+        self.thread.join()
+        self.thread = None
+
+    def _stream(self):
+        chunk = 512  # 320 at 16000 Hz is 20ms, 512 is required for Silero-VAD
+
+        # Docs say this only raises an error if neither `input` nor `output` is True
+        stream = self.audio.open(
+            format=pyaudio.paFloat32,
+            channels=1,
+            rate=16000,
+            input=True,
+            input_device_index=self.microphone["index"],
+            frames_per_buffer=chunk,
+        )
+
+        try:
+            while not state.exit_event.is_set():
+                data = stream.read(chunk)
+                self.socket.send(data)
+        except IOError as e:
+            logger.error("Stopped listening: failed to get audio from microphone.", exc_info=e)
+        finally:
+            stream.stop_stream()
+            stream.close()

src/robot_interface/endpoints/video_sender.py

@@ -1,6 +1,5 @@
 import zmq
 import threading
-import qi
 import logging
 
 from robot_interface.endpoints.socket_base import SocketBase

src/robot_interface/main.py

@@ -1,4 +1,7 @@
 import logging
+
+from robot_interface.endpoints.audio_sender import AudioSender
+
 logging.basicConfig(level=logging.DEBUG)
 
 import zmq
@@ -25,8 +28,11 @@ def main_loop(context):
 
     video_sender = VideoSender(context)
     state.sockets.append(video_sender)
+    audio_sender = AudioSender(context)
+    state.sockets.append(audio_sender)
 
     video_sender.start_video_rcv()
+    audio_sender.start()
 
     # Sockets that can run on the main thread. These sockets' endpoints should not block for long (say 50 ms at most).
     receivers = [main_receiver, actuation_receiver]

src/robot_interface/utils/microphone.py

@@ -0,0 +1,120 @@
+from __future__ import unicode_literals  # So that `print` can print Unicode characters in names
+import logging
+import sys
+
+logger = logging.getLogger(__name__)
+
+
+def get_microphones(audio):
+    """
+    Get audio devices which have input channels.
+
+    :param audio: An instance of PyAudio to use.
+    :type audio: pyaudio.PyAudio
+
+    :return: An interator of PaAudio dicts containing information about the microphone devices.
+    :rtype: Iterator[dict]
+    """
+    for i in range(audio.get_device_count()):
+        device = audio.get_device_info_by_index(i)
+        if device["maxInputChannels"] > 0:
+            yield device
+
+
+def choose_mic_interactive(audio):
+    """
+    Choose a microphone to use, interactively in the CLI.
+
+    :param audio: An instance of PyAudio to use.
+    :type audio: pyaudio.PyAudio
+
+    :return: A dictionary from PyAudio containing information about the microphone to use, or None
+    if there is no microphone.
+    :rtype: dict | None
+    """
+    microphones = list(get_microphones(audio))
+    if len(microphones) == 0: return None
+
+    print("Found {} microphones:".format(len(microphones)))
+    for i, mic in enumerate(microphones):
+        print("- {}: {}".format(i, mic["name"]))
+
+    chosen_microphone = None
+    while chosen_microphone is None:
+        chosen = raw_input("Which device would you like to use?\n> ")
+        try:
+            chosen = int(chosen)
+            if chosen < 0 or chosen >= len(microphones): raise ValueError()
+            chosen_microphone = microphones[chosen]
+        except ValueError:
+            print("Please enter a number between 0 and {}".format(len(microphones)-1))
+
+    logger.info("Chose microphone \"{}\"".format(chosen_microphone["name"]))
+    return chosen_microphone
+
+
+def choose_mic_default(audio):
+    """
+    Get the system's default microphone to use.
+
+    :param audio: An instance of PyAudio to use.
+    :type audio: pyaudio.PyAudio
+
+    :return: A dictionary from PyAudio containing information about the microphone to use, or None
+    if there is no microphone.
+    :rtype: dict | None
+    """
+    try:
+        return audio.get_default_input_device_info()
+    except IOError:
+        return None
+
+
+def choose_mic_arguments(audio):
+    """
+    Get a microphone to use from command line arguments.
+
+    :param audio: An instance of PyAudio to use.
+    :type audio: pyaudio.PyAudio
+
+    :return: A dictionary from PyAudio containing information about the microphone to use, or None
+    if there is no microphone satisfied by the arguments.
+    :rtype: dict | None
+    """
+    microphone_name = None
+    for i, arg in enumerate(sys.argv):
+        if arg == "--microphone" and len(sys.argv) > i+1:
+            microphone_name = sys.argv[i+1].strip()
+        if arg.startswith("--microphone="):
+            microphone_name = arg[13:].strip()
+
+    if not microphone_name: return None
+
+    available_mics = list(get_microphones(audio))
+    for mic in available_mics:
+        if mic["name"] == microphone_name:
+            return mic
+
+    available_mic_names = [mic["name"] for mic in available_mics]
+    logger.warning("Microphone \"{}\" not found. Choose one of {}"
+                   .format(microphone_name, available_mic_names))
+
+    return None
+
+
+def choose_mic(audio):
+    """
+    Get a microphone to use. Firstly, tries to see if there's an application argument specifying the
+    microphone to use. If not, get the default microphone.
+
+    :param audio: An instance of PyAudio to use.
+    :type audio: pyaudio.PyAudio
+
+    :return: A dictionary from PyAudio containing information about the microphone to use, or None
+    if there is no microphone.
+    :rtype: dict | None
+    """
+    chosen_mic = choose_mic_arguments(audio)
+    if chosen_mic: return chosen_mic
+
+    return choose_mic_default(audio)

test/common/microphone_utils.py

@@ -0,0 +1,152 @@
+from __future__ import unicode_literals  # So that we can format strings with Unicode characters
+import random
+import sys
+from StringIO import StringIO
+
+from robot_interface.utils.microphone import (
+    choose_mic_default,
+    choose_mic_interactive,
+    choose_mic_arguments,
+    choose_mic,
+    get_microphones,
+)
+
+
+class MicrophoneUtils(object):
+    """Shared tests for any PyAudio-like implementation, e.g. mock and real."""
+
+    def test_choose_mic_default(self, pyaudio_instance):
+        """
+        The result must contain at least "index", as this is used to identify the microphone.
+        The "name" is used for logging, so it should also exist.
+        It must have one or more channels.
+        Lastly it must be capable of sending at least 16000 samples per second.
+        """
+        result = choose_mic_default(pyaudio_instance)
+        assert "index" in result
+        assert isinstance(result["index"], (int, long))
+
+        assert "name" in result
+        assert isinstance(result["name"], (str, unicode))
+
+        assert "maxInputChannels" in result
+        assert isinstance(result["maxInputChannels"], (int, long))
+        assert result["maxInputChannels"] > 0
+
+        assert "defaultSampleRate" in result
+        assert isinstance(result["defaultSampleRate"], float)
+        assert result["defaultSampleRate"] >= 16000
+
+    def test_choose_mic_interactive_input_not_int(self, pyaudio_instance, mocker):
+        """
+        First mock an input that's not an integer, then a valid integer. There should be no errors.
+        """
+        mock_input = mocker.patch("__builtin__.raw_input", side_effect=["not an integer", "0"])
+        fake_out = StringIO()
+        mocker.patch.object(sys, "stdout", fake_out)
+
+        result = choose_mic_interactive(pyaudio_instance)
+        assert "index" in result
+        assert isinstance(result["index"], (int, long))
+        assert result["index"] == 0
+
+        assert mock_input.called
+
+        assert any(p.startswith("Please enter a number") for p in fake_out.getvalue().splitlines())
+
+    def test_choose_mic_interactive_negative_index(self, pyaudio_instance, mocker):
+        """
+        Make sure that the interactive method does not allow negative integers as input.
+        """
+        mock_input = mocker.patch("__builtin__.raw_input", side_effect=["-1", "0"])
+        fake_out = StringIO()
+        mocker.patch.object(sys, "stdout", fake_out)
+
+        result = choose_mic_interactive(pyaudio_instance)
+        assert "index" in result
+        assert isinstance(result["index"], (int, long))
+        assert result["index"] == 0
+
+        assert mock_input.called
+
+        assert any(p.startswith("Please enter a number") for p in fake_out.getvalue().splitlines())
+
+    def test_choose_mic_interactive_index_too_high(self, pyaudio_instance, mocker):
+        """
+        Make sure that the interactive method does not allow indices higher than the highest mic index.
+        """
+        real_count = len(list(get_microphones(pyaudio_instance)))
+        mock_input = mocker.patch("__builtin__.raw_input", side_effect=[str(real_count), "0"])
+        fake_out = StringIO()
+        mocker.patch.object(sys, "stdout", fake_out)
+
+        result = choose_mic_interactive(pyaudio_instance)
+        assert "index" in result
+        assert isinstance(result["index"], (int, long))
+
+        assert mock_input.called
+
+        assert any(p.startswith("Please enter a number") for p in fake_out.getvalue().splitlines())
+
+    def test_choose_mic_interactive_random_index(self, pyaudio_instance, mocker):
+        """
+        Get a random index from the list of available mics, make sure it's correct.
+        """
+        microphones = list(get_microphones(pyaudio_instance))
+        random_index = random.randrange(len(microphones))
+        mocker.patch("__builtin__.raw_input", side_effect=[str(random_index)])
+
+        result = choose_mic_interactive(pyaudio_instance)
+        assert "index" in result
+        assert isinstance(result["index"], (int, long))
+        assert result["index"] == microphones[random_index]["index"]
+
+    def test_choose_mic_no_arguments(self, pyaudio_instance, mocker):
+        mocker.patch.object(sys, "argv", [])
+
+        result = choose_mic_arguments(pyaudio_instance)
+
+        assert result is None
+
+    def test_choose_mic_arguments(self, pyaudio_instance, mocker):
+        for mic in get_microphones(pyaudio_instance):
+            mocker.patch.object(sys, "argv", ["--microphone", mic["name"]])
+
+            result = choose_mic_arguments(pyaudio_instance)
+
+            assert result is not None
+            assert result == mic
+
+    def test_choose_mic_arguments_eq(self, pyaudio_instance, mocker):
+        for mic in get_microphones(pyaudio_instance):
+            mocker.patch.object(sys, "argv", ["--microphone={}".format(mic["name"])])
+
+            result = choose_mic_arguments(pyaudio_instance)
+
+            assert result is not None
+            assert result == mic
+
+    def test_choose_mic_arguments_not_exits(self, pyaudio_instance, mocker):
+        mocker.patch.object(sys, "argv", ["--microphone", "Surely this microphone doesn't exist"])
+
+        result = choose_mic_arguments(pyaudio_instance)
+
+        assert result is None
+
+    def test_choose_mic_with_argument(self, pyaudio_instance, mocker):
+        mic = next(get_microphones(pyaudio_instance))
+        mocker.patch.object(sys, "argv", ["--microphone", mic["name"]])
+
+        result = choose_mic(pyaudio_instance)
+
+        assert result is not None
+        assert result == mic
+
+    def test_choose_mic_no_argument(self, pyaudio_instance, mocker):
+        default_mic = choose_mic_default(pyaudio_instance)
+        mocker.patch.object(sys, "argv", [])
+
+        result = choose_mic(pyaudio_instance)
+
+        assert result is not None
+        assert result == default_mic

test/integration/test_microphone_utils.py

@@ -0,0 +1,20 @@
+import pyaudio
+
+import pytest
+
+from common.microphone_utils import MicrophoneUtils
+
+
+@pytest.fixture
+def pyaudio_instance():
+    audio = pyaudio.PyAudio()
+    try:
+        audio.get_default_input_device_info()
+        return audio
+    except IOError:
+        pytest.skip("No microphone available to test with.")
+
+
+class TestAudioIntegration(MicrophoneUtils):
+    """Run shared audio behavior tests with the mock implementation."""
+    pass

test/unit/test_audio_sender.py

@@ -0,0 +1,104 @@
+# coding=utf-8
+import os
+import time
+
+import mock
+import pytest
+import zmq
+
+from robot_interface.endpoints.audio_sender import AudioSender
+
+
+@pytest.fixture
+def zmq_context():
+    context = zmq.Context()
+    yield context
+
+
+def test_no_microphone(zmq_context, mocker):
+    mock_info_logger = mocker.patch("robot_interface.endpoints.audio_sender.logger.info")
+    mock_choose_mic = mocker.patch("robot_interface.endpoints.audio_sender.choose_mic")
+    mock_choose_mic.return_value = None
+
+    sender = AudioSender(zmq_context)
+    assert sender.microphone is None
+
+    sender.start()
+    assert sender.thread is None
+    mock_info_logger.assert_called()
+
+    sender.wait_until_done()  # Should return early because we didn't start a thread
+
+
+def test_unicode_mic_name(zmq_context, mocker):
+    mocker.patch("robot_interface.endpoints.audio_sender.threading")
+    mock_choose_mic = mocker.patch("robot_interface.endpoints.audio_sender.choose_mic")
+    mock_choose_mic.return_value = {"name": u"• Some Unicode name"}
+
+    sender = AudioSender(zmq_context)
+    assert sender.microphone is not None
+
+    # `.start()` logs the name of the microphone. It should not give an error if it contains Unicode
+    #  symbols.
+    sender.start()
+    assert sender.thread is not None
+
+    sender.wait_until_done()  # Should return instantly because we didn't start a real thread
+
+
+def _fake_read(num_frames):
+    return os.urandom(num_frames * 4)
+
+
+def test_sending_audio(mocker):
+    mock_choose_mic = mocker.patch("robot_interface.endpoints.audio_sender.choose_mic")
+    mock_choose_mic.return_value = {"name": u"Some mic", "index": 0L}
+
+    mock_state = mocker.patch("robot_interface.endpoints.audio_sender.state")
+    mock_state.exit_event.is_set.side_effect = [False, True]
+
+    mock_zmq_context = mock.Mock()
+    send_socket = mock.Mock()
+
+    # If there's something wrong with the microphone, it will raise an IOError when `read`ing.
+    stream = mock.Mock()
+    stream.read = _fake_read
+
+    sender = AudioSender(mock_zmq_context)
+    sender.socket.send = send_socket
+    sender.audio.open = mock.Mock()
+    sender.audio.open.return_value = stream
+
+    sender.start()
+    sender.wait_until_done()
+
+    send_socket.assert_called()
+
+
+def _fake_read_error(num_frames):
+    raise IOError()
+
+
+def test_break_microphone(mocker):
+    mock_choose_mic = mocker.patch("robot_interface.endpoints.audio_sender.choose_mic")
+    mock_choose_mic.return_value = {"name": u"Some mic", "index": 0L}
+
+    mock_state = mocker.patch("robot_interface.endpoints.audio_sender.state")
+    mock_state.exit_event.is_set.side_effect = [False, True]
+
+    mock_zmq_context = mock.Mock()
+    send_socket = mock.Mock()
+
+    # If there's something wrong with the microphone, it will raise an IOError when `read`ing.
+    stream = mock.Mock()
+    stream.read = _fake_read_error
+
+    sender = AudioSender(mock_zmq_context)
+    sender.socket.send = send_socket
+    sender.audio.open = mock.Mock()
+    sender.audio.open.return_value = stream
+
+    sender.start()
+    sender.wait_until_done()
+
+    send_socket.assert_not_called()

test/unit/test_microphone_utils.py

@@ -0,0 +1,85 @@
+# coding=utf-8
+import mock
+import pytest
+
+from common.microphone_utils import MicrophoneUtils
+from robot_interface.utils.microphone import choose_mic_default, choose_mic_interactive
+
+
+class MockPyAudio:
+    def __init__(self):
+        # You can predefine fake device info here
+        self.devices = [
+            {
+                "index": 0,
+                "name": u"Someone’s Microphone",  # Using a Unicode ’ character
+                "maxInputChannels": 2,
+                "maxOutputChannels": 0,
+                "defaultSampleRate": 44100.0,
+                "defaultLowInputLatency": 0.01,
+                "defaultLowOutputLatency": 0.01,
+                "defaultHighInputLatency": 0.1,
+                "defaultHighOutputLatency": 0.1,
+                "hostApi": 0,
+            },
+            {
+                "index": 1,
+                "name": u"Mock Speaker 1",
+                "maxInputChannels": 0,
+                "maxOutputChannels": 2,
+                "defaultSampleRate": 48000.0,
+                "defaultLowInputLatency": 0.01,
+                "defaultLowOutputLatency": 0.01,
+                "defaultHighInputLatency": 0.1,
+                "defaultHighOutputLatency": 0.1,
+                "hostApi": 0,
+            },
+        ]
+
+    def get_device_count(self):
+        """Return the number of available mock devices."""
+        return len(self.devices)
+
+    def get_device_info_by_index(self, index):
+        """Return information for a given mock device index."""
+        if 0 <= index < len(self.devices):
+            return self.devices[index]
+        else:
+            raise IOError("Invalid device index: {}".format(index))
+
+    def get_default_input_device_info(self):
+        """Return info for a default mock input device."""
+        for device in self.devices:
+            if device.get("maxInputChannels", 0) > 0:
+                return device
+        raise IOError("No default input device found")
+
+
+@pytest.fixture
+def pyaudio_instance():
+    return MockPyAudio()
+
+
+def _raise_io_error():
+    raise IOError()
+
+
+class TestAudioUnit(MicrophoneUtils):
+    """Run shared audio behavior tests with the mock implementation."""
+    def test_choose_mic_default_no_mic(self):
+        mock_pyaudio = mock.Mock()
+        mock_pyaudio.get_device_count = mock.Mock(return_value=0L)
+        mock_pyaudio.get_default_input_device_info = _raise_io_error
+
+        result = choose_mic_default(mock_pyaudio)
+
+        assert result is None
+
+    def test_choose_mic_interactive_no_mic(self):
+        mock_pyaudio = mock.Mock()
+        mock_pyaudio.get_device_count = mock.Mock(return_value=0L)
+        mock_pyaudio.get_default_input_device_info = _raise_io_error
+
+        result = choose_mic_interactive(mock_pyaudio)
+
+        assert result is None