add offline websocket server/client (#98)

python-api-examples/offline-websocket-client-decode-files-paralell.py

@@ -0,0 +1,158 @@
+#!/usr/bin/env python3
+#
+# Copyright (c)  2023  Xiaomi Corporation
+
+"""
+A websocket client for sherpa-onnx-offline-websocket-server
+
+This file shows how to transcribe multiple
+files in parallel. We create a separate connection for transcribing each file.
+
+Usage:
+    ./offline-websocket-client-decode-files-parallel.py \
+      --server-addr localhost \
+      --server-port 6006 \
+      /path/to/foo.wav \
+      /path/to/bar.wav \
+      /path/to/16kHz.wav \
+      /path/to/8kHz.wav
+
+(Note: You have to first start the server before starting the client)
+
+You can find the server at
+https://github.com/k2-fsa/sherpa-onnx/blob/master/sherpa-onnx/csrc/offline-websocket-server.cc
+
+Note: The server is implemented in C++.
+"""
+
+import argparse
+import asyncio
+import logging
+import wave
+from typing import Tuple
+
+try:
+    import websockets
+except ImportError:
+    print("please run:")
+    print("")
+    print("  pip install websockets")
+    print("")
+    print("before you run this script")
+    print("")
+
+import numpy as np
+
+
+def get_args():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    parser.add_argument(
+        "--server-addr",
+        type=str,
+        default="localhost",
+        help="Address of the server",
+    )
+
+    parser.add_argument(
+        "--server-port",
+        type=int,
+        default=6006,
+        help="Port of the server",
+    )
+
+    parser.add_argument(
+        "sound_files",
+        type=str,
+        nargs="+",
+        help="The input sound file(s) to decode. Each file must be of WAVE"
+        "format with a single channel, and each sample has 16-bit, "
+        "i.e., int16_t. "
+        "The sample rate of the file can be arbitrary and does not need to "
+        "be 16 kHz",
+    )
+
+    return parser.parse_args()
+
+
+def read_wave(wave_filename: str) -> Tuple[np.ndarray, int]:
+    """
+    Args:
+      wave_filename:
+        Path to a wave file. It should be single channel and each sample should
+        be 16-bit. Its sample rate does not need to be 16kHz.
+    Returns:
+      Return a tuple containing:
+       - A 1-D array of dtype np.float32 containing the samples, which are
+       normalized to the range [-1, 1].
+       - sample rate of the wave file
+    """
+
+    with wave.open(wave_filename) as f:
+        assert f.getnchannels() == 1, f.getnchannels()
+        assert f.getsampwidth() == 2, f.getsampwidth()  # it is in bytes
+        num_samples = f.getnframes()
+        samples = f.readframes(num_samples)
+        samples_int16 = np.frombuffer(samples, dtype=np.int16)
+        samples_float32 = samples_int16.astype(np.float32)
+
+        samples_float32 = samples_float32 / 32768
+        return samples_float32, f.getframerate()
+
+
+async def run(
+    server_addr: str,
+    server_port: int,
+    wave_filename: str,
+):
+    async with websockets.connect(
+        f"ws://{server_addr}:{server_port}"
+    ) as websocket:  # noqa
+        logging.info(f"Sending {wave_filename}")
+        samples, sample_rate = read_wave(wave_filename)
+        assert isinstance(sample_rate, int)
+        assert samples.dtype == np.float32, samples.dtype
+        assert samples.ndim == 1, samples.dim
+        buf = sample_rate.to_bytes(4, byteorder="little")  # 4 bytes
+        buf += (samples.size * 4).to_bytes(4, byteorder="little")
+        buf += samples.tobytes()
+
+        await websocket.send(buf)
+
+        decoding_results = await websocket.recv()
+        logging.info(f"{wave_filename}\n{decoding_results}")
+
+        # to signal that the client has sent all the data
+        await websocket.send("Done")
+
+
+async def main():
+    args = get_args()
+    logging.info(vars(args))
+
+    server_addr = args.server_addr
+    server_port = args.server_port
+    sound_files = args.sound_files
+
+    all_tasks = []
+    for wave_filename in sound_files:
+        task = asyncio.create_task(
+            run(
+                server_addr=server_addr,
+                server_port=server_port,
+                wave_filename=wave_filename,
+            )
+        )
+        all_tasks.append(task)
+
+    await asyncio.gather(*all_tasks)
+
+
+if __name__ == "__main__":
+    formatter = (
+        "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"  # noqa
+    )
+    logging.basicConfig(format=formatter, level=logging.INFO)
+    asyncio.run(main())

python-api-examples/offline-websocket-client-decode-files-sequential.py

@@ -0,0 +1,152 @@
+#!/usr/bin/env python3
+#
+# Copyright (c)  2023  Xiaomi Corporation
+
+"""
+A websocket client for sherpa-onnx-offline-websocket-server
+
+This file shows how to use a single connection to transcribe multiple
+files sequentially.
+
+Usage:
+    ./offline-websocket-client-decode-files-sequential.py \
+      --server-addr localhost \
+      --server-port 6006 \
+      /path/to/foo.wav \
+      /path/to/bar.wav \
+      /path/to/16kHz.wav \
+      /path/to/8kHz.wav
+
+(Note: You have to first start the server before starting the client)
+
+You can find the server at
+https://github.com/k2-fsa/sherpa-onnx/blob/master/sherpa-onnx/csrc/offline-websocket-server.cc
+
+Note: The server is implemented in C++.
+"""
+
+import argparse
+import asyncio
+import logging
+import wave
+from typing import List, Tuple
+
+try:
+    import websockets
+except ImportError:
+    print("please run:")
+    print("")
+    print("  pip install websockets")
+    print("")
+    print("before you run this script")
+    print("")
+
+import numpy as np
+
+
+def get_args():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    parser.add_argument(
+        "--server-addr",
+        type=str,
+        default="localhost",
+        help="Address of the server",
+    )
+
+    parser.add_argument(
+        "--server-port",
+        type=int,
+        default=6006,
+        help="Port of the server",
+    )
+
+    parser.add_argument(
+        "sound_files",
+        type=str,
+        nargs="+",
+        help="The input sound file(s) to decode. Each file must be of WAVE"
+        "format with a single channel, and each sample has 16-bit, "
+        "i.e., int16_t. "
+        "The sample rate of the file can be arbitrary and does not need to "
+        "be 16 kHz",
+    )
+
+    return parser.parse_args()
+
+
+def read_wave(wave_filename: str) -> Tuple[np.ndarray, int]:
+    """
+    Args:
+      wave_filename:
+        Path to a wave file. It should be single channel and each sample should
+        be 16-bit. Its sample rate does not need to be 16kHz.
+    Returns:
+      Return a tuple containing:
+       - A 1-D array of dtype np.float32 containing the samples, which are
+       normalized to the range [-1, 1].
+       - sample rate of the wave file
+    """
+
+    with wave.open(wave_filename) as f:
+        assert f.getnchannels() == 1, f.getnchannels()
+        assert f.getsampwidth() == 2, f.getsampwidth()  # it is in bytes
+        num_samples = f.getnframes()
+        samples = f.readframes(num_samples)
+        samples_int16 = np.frombuffer(samples, dtype=np.int16)
+        samples_float32 = samples_int16.astype(np.float32)
+
+        samples_float32 = samples_float32 / 32768
+        return samples_float32, f.getframerate()
+
+
+async def run(
+    server_addr: str,
+    server_port: int,
+    sound_files: List[str],
+):
+    async with websockets.connect(
+        f"ws://{server_addr}:{server_port}"
+    ) as websocket:  # noqa
+        for wave_filename in sound_files:
+            logging.info(f"Sending {wave_filename}")
+            samples, sample_rate = read_wave(wave_filename)
+            assert isinstance(sample_rate, int)
+            assert samples.dtype == np.float32, samples.dtype
+            assert samples.ndim == 1, samples.dim
+            buf = sample_rate.to_bytes(4, byteorder="little")  # 4 bytes
+            buf += (samples.size * 4).to_bytes(4, byteorder="little")
+            buf += samples.tobytes()
+
+            await websocket.send(buf)
+
+            decoding_results = await websocket.recv()
+            print(decoding_results)
+
+        # to signal that the client has sent all the data
+        await websocket.send("Done")
+
+
+async def main():
+    args = get_args()
+    logging.info(vars(args))
+
+    server_addr = args.server_addr
+    server_port = args.server_port
+    sound_files = args.sound_files
+
+    await run(
+        server_addr=server_addr,
+        server_port=server_port,
+        sound_files=sound_files,
+    )
+
+
+if __name__ == "__main__":
+    formatter = (
+        "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"  # noqa
+    )
+    logging.basicConfig(format=formatter, level=logging.INFO)
+    asyncio.run(main())

python-api-examples/online-decode-files.py

@@ -1,8 +1,15 @@
 #!/usr/bin/env python3
 
 """
-This file demonstrates how to use sherpa-onnx Python API to recognize
-a single file.
+This file demonstrates how to use sherpa-onnx Python API to transcribe
+file(s) with a streaming model.
+
+Usage:
+    ./online-decode-files.py \
+      /path/to/foo.wav \
+      /path/to/bar.wav \
+      /path/to/16kHz.wav \
+      /path/to/8kHz.wav
 
 Please refer to
 https://k2-fsa.github.io/sherpa/onnx/index.html
@@ -13,17 +20,12 @@ import argparse
 import time
 import wave
 from pathlib import Path
+from typing import Tuple
 
 import numpy as np
 import sherpa_onnx
 
 
-def assert_file_exists(filename: str):
-    assert Path(
-        filename
-    ).is_file(), f"{filename} does not exist!\nPlease 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
@@ -68,26 +70,58 @@ def get_args():
     )
 
     parser.add_argument(
-        "--wave-filename",
+        "sound_files",
         type=str,
-        help="""Path to the wave filename.
-        Should have a single channel with 16-bit samples.
-        It does not need to be 16kHz. It can have any sampling rate.
-        """,
+        nargs="+",
+        help="The input sound file(s) to decode. Each file must be of WAVE"
+        "format with a single channel, and each sample has 16-bit, "
+        "i.e., int16_t. "
+        "The sample rate of the file can be arbitrary and does not need to "
+        "be 16 kHz",
     )
 
     return parser.parse_args()
 
 
+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 read_wave(wave_filename: str) -> Tuple[np.ndarray, int]:
+    """
+    Args:
+      wave_filename:
+        Path to a wave file. It should be single channel and each sample should
+        be 16-bit. Its sample rate does not need to be 16kHz.
+    Returns:
+      Return a tuple containing:
+       - A 1-D array of dtype np.float32 containing the samples, which are
+       normalized to the range [-1, 1].
+       - sample rate of the wave file
+    """
+
+    with wave.open(wave_filename) as f:
+        assert f.getnchannels() == 1, f.getnchannels()
+        assert f.getsampwidth() == 2, f.getsampwidth()  # it is in bytes
+        num_samples = f.getnframes()
+        samples = f.readframes(num_samples)
+        samples_int16 = np.frombuffer(samples, dtype=np.int16)
+        samples_float32 = samples_int16.astype(np.float32)
+
+        samples_float32 = samples_float32 / 32768
+        return samples_float32, f.getframerate()
+
+
 def main():
     args = get_args()
     assert_file_exists(args.encoder)
     assert_file_exists(args.decoder)
     assert_file_exists(args.joiner)
     assert_file_exists(args.tokens)
-    if not Path(args.wave_filename).is_file():
-        print(f"{args.wave_filename} does not exist!")
-        return
 
     recognizer = sherpa_onnx.OnlineRecognizer(
         tokens=args.tokens,
@@ -99,42 +133,44 @@ def main():
         feature_dim=80,
         decoding_method=args.decoding_method,
     )
-    with wave.open(args.wave_filename) as f:
-        # If the wave file has a different sampling rate from the one
-        # expected by the model (16 kHz in our case), we will do
-        # resampling inside sherpa-onnx
-        wave_file_sample_rate = f.getframerate()
 
-        assert f.getnchannels() == 1, f.getnchannels()
-        assert f.getsampwidth() == 2, f.getsampwidth()  # it is in bytes
-        num_samples = f.getnframes()
-        samples = f.readframes(num_samples)
-        samples_int16 = np.frombuffer(samples, dtype=np.int16)
-        samples_float32 = samples_int16.astype(np.float32)
-
-        samples_float32 = samples_float32 / 32768
-
-    duration = len(samples_float32) / wave_file_sample_rate
-
-    start_time = time.time()
     print("Started!")
+    start_time = time.time()
 
-    stream = recognizer.create_stream()
+    streams = []
+    total_duration = 0
+    for wave_filename in args.sound_files:
+        assert_file_exists(wave_filename)
+        samples, sample_rate = read_wave(wave_filename)
+        duration = len(samples) / sample_rate
+        total_duration += duration
 
-    stream.accept_waveform(wave_file_sample_rate, samples_float32)
+        s = recognizer.create_stream()
+        s.accept_waveform(sample_rate, samples)
 
-    tail_paddings = np.zeros(int(0.2 * wave_file_sample_rate), dtype=np.float32)
-    stream.accept_waveform(wave_file_sample_rate, tail_paddings)
+        tail_paddings = np.zeros(int(0.2 * sample_rate), dtype=np.float32)
+        s.accept_waveform(sample_rate, tail_paddings)
 
-    stream.input_finished()
+        s.input_finished()
 
-    while recognizer.is_ready(stream):
-        recognizer.decode_stream(stream)
+        streams.append(s)
 
-    print(recognizer.get_result(stream))
-
-    print("Done!")
+    while True:
+        ready_list = []
+        for s in streams:
+            if recognizer.is_ready(s):
+                ready_list.append(s)
+        if len(ready_list) == 0:
+            break
+        recognizer.decode_streams(ready_list)
+    results = [recognizer.get_result(s) for s in streams]
     end_time = time.time()
+    print("Done!")
+
+    for wave_filename, result in zip(args.sound_files, results):
+        print(f"{wave_filename}\n{result}")
+        print("-" * 10)
+
     elapsed_seconds = end_time - start_time
     rtf = elapsed_seconds / duration
     print(f"num_threads: {args.num_threads}")

python-api-examples/online-websocket-client-decode-file.py

@@ -27,7 +27,6 @@ https://github.com/k2-fsa/sherpa-onnx/blob/master/sherpa-onnx/csrc/online-websoc
 import argparse
 import asyncio
 import logging
-import time
 import wave
 
 try:

python-api-examples/online-websocket-client-microphone.py

@@ -24,13 +24,12 @@ https://github.com/k2-fsa/sherpa-onnx/blob/master/sherpa-onnx/csrc/online-websoc
 import argparse
 import asyncio
 import sys
-import time
 
 import numpy as np
 
 try:
     import sounddevice as sd
-except ImportError as e:
+except ImportError:
     print("Please install sounddevice first. You can use")
     print()
     print("  pip install sounddevice")
@@ -134,7 +133,7 @@ async def run(
             await websocket.send(indata.tobytes())
 
         decoding_results = await receive_task
-        print("\nFinal result is:\n{decoding_results}")
+        print(f"\nFinal result is:\n{decoding_results}")
 
 
 async def main():

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

@@ -13,7 +13,7 @@ from pathlib import Path
 
 try:
     import sounddevice as sd
-except ImportError as e:
+except ImportError:
     print("Please install sounddevice first. You can use")
     print()
     print("  pip install sounddevice")
@@ -25,9 +25,11 @@ import sherpa_onnx
 
 
 def assert_file_exists(filename: str):
-    assert Path(
-        filename
-    ).is_file(), f"{filename} does not exist!\nPlease refer to https://k2-fsa.github.io/sherpa/onnx/pretrained_models/index.html to download it"
+    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():

python-api-examples/speech-recognition-from-microphone.py

@@ -12,7 +12,7 @@ from pathlib import Path
 
 try:
     import sounddevice as sd
-except ImportError as e:
+except ImportError:
     print("Please install sounddevice first. You can use")
     print()
     print("  pip install sounddevice")
@@ -24,9 +24,11 @@ import sherpa_onnx
 
 
 def assert_file_exists(filename: str):
-    assert Path(
-        filename
-    ).is_file(), f"{filename} does not exist!\nPlease refer to https://k2-fsa.github.io/sherpa/onnx/pretrained_models/index.html to download it"
+    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():