EngineX-Iluvatar/enginex_bi_series-sherpa-onnx
Archived
16
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Add C++ runtime and Python APIs for Moonshine models (#1473)

Browse Source
This commit is contained in:
Fangjun Kuang
2024-10-26 14:34:07 +08:00
committed by GitHub
parent 0f2732e4e8
commit 669f5ef441
33 changed files with 1572 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files

117
python-api-examples/generate-subtitles.py
View File

@@ -47,7 +47,19 @@ wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/silero_v
--feature-dim=80 \
/path/to/test.mp4
(3) For Whisper models
(3) For Moonshine models
./python-api-examples/generate-subtitles.py \
--silero-vad-model=/path/to/silero_vad.onnx \
--moonshine-preprocessor=./sherpa-onnx-moonshine-tiny-en-int8/preprocess.onnx \
--moonshine-encoder=./sherpa-onnx-moonshine-tiny-en-int8/encode.int8.onnx \
--moonshine-uncached-decoder=./sherpa-onnx-moonshine-tiny-en-int8/uncached_decode.int8.onnx \
--moonshine-cached-decoder=./sherpa-onnx-moonshine-tiny-en-int8/cached_decode.int8.onnx \
--tokens=./sherpa-onnx-moonshine-tiny-en-int8/tokens.txt \
--num-threads=2 \
/path/to/test.mp4
(4) For Whisper models
./python-api-examples/generate-subtitles.py \
--silero-vad-model=/path/to/silero_vad.onnx \
@@ -58,7 +70,7 @@ wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/silero_v
--num-threads=2 \
/path/to/test.mp4
(4) For SenseVoice CTC models
(5) For SenseVoice CTC models
./python-api-examples/generate-subtitles.py \
--silero-vad-model=/path/to/silero_vad.onnx \
@@ -68,7 +80,7 @@ wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/silero_v
/path/to/test.mp4
(5) For WeNet CTC models
(6) For WeNet CTC models
./python-api-examples/generate-subtitles.py \
--silero-vad-model=/path/to/silero_vad.onnx \
@@ -83,6 +95,7 @@ to install sherpa-onnx and to download non-streaming pre-trained models
used in this file.
"""
import argparse
import datetime as dt
import shutil
import subprocess
import sys
@@ -157,7 +170,7 @@ def get_args():
parser.add_argument(
"--num-threads",
type=int,
default=1,
default=2,
help="Number of threads for neural network computation",
)
@@ -208,6 +221,34 @@ def get_args():
""",
)
parser.add_argument(
"--moonshine-preprocessor",
default="",
type=str,
help="Path to moonshine preprocessor model",
)
parser.add_argument(
"--moonshine-encoder",
default="",
type=str,
help="Path to moonshine encoder model",
)
parser.add_argument(
"--moonshine-uncached-decoder",
default="",
type=str,
help="Path to moonshine uncached decoder model",
)
parser.add_argument(
"--moonshine-cached-decoder",
default="",
type=str,
help="Path to moonshine cached decoder model",
)
parser.add_argument(
"--decoding-method",
type=str,
@@ -263,6 +304,12 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
assert len(args.wenet_ctc) == 0, args.wenet_ctc
assert len(args.whisper_encoder) == 0, args.whisper_encoder
assert len(args.whisper_decoder) == 0, args.whisper_decoder
assert len(args.moonshine_preprocessor) == 0, args.moonshine_preprocessor
assert len(args.moonshine_encoder) == 0, args.moonshine_encoder
assert (
len(args.moonshine_uncached_decoder) == 0
), args.moonshine_uncached_decoder
assert len(args.moonshine_cached_decoder) == 0, args.moonshine_cached_decoder
assert_file_exists(args.encoder)
assert_file_exists(args.decoder)
@@ -284,6 +331,12 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
assert len(args.wenet_ctc) == 0, args.wenet_ctc
assert len(args.whisper_encoder) == 0, args.whisper_encoder
assert len(args.whisper_decoder) == 0, args.whisper_decoder
assert len(args.moonshine_preprocessor) == 0, args.moonshine_preprocessor
assert len(args.moonshine_encoder) == 0, args.moonshine_encoder
assert (
len(args.moonshine_uncached_decoder) == 0
), args.moonshine_uncached_decoder
assert len(args.moonshine_cached_decoder) == 0, args.moonshine_cached_decoder
assert_file_exists(args.paraformer)
@@ -300,6 +353,12 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
assert len(args.wenet_ctc) == 0, args.wenet_ctc
assert len(args.whisper_encoder) == 0, args.whisper_encoder
assert len(args.whisper_decoder) == 0, args.whisper_decoder
assert len(args.moonshine_preprocessor) == 0, args.moonshine_preprocessor
assert len(args.moonshine_encoder) == 0, args.moonshine_encoder
assert (
len(args.moonshine_uncached_decoder) == 0
), args.moonshine_uncached_decoder
assert len(args.moonshine_cached_decoder) == 0, args.moonshine_cached_decoder
assert_file_exists(args.sense_voice)
recognizer = sherpa_onnx.OfflineRecognizer.from_sense_voice(
@@ -312,6 +371,12 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
elif args.wenet_ctc:
assert len(args.whisper_encoder) == 0, args.whisper_encoder
assert len(args.whisper_decoder) == 0, args.whisper_decoder
assert len(args.moonshine_preprocessor) == 0, args.moonshine_preprocessor
assert len(args.moonshine_encoder) == 0, args.moonshine_encoder
assert (
len(args.moonshine_uncached_decoder) == 0
), args.moonshine_uncached_decoder
assert len(args.moonshine_cached_decoder) == 0, args.moonshine_cached_decoder
assert_file_exists(args.wenet_ctc)
@@ -327,6 +392,12 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
elif args.whisper_encoder:
assert_file_exists(args.whisper_encoder)
assert_file_exists(args.whisper_decoder)
assert len(args.moonshine_preprocessor) == 0, args.moonshine_preprocessor
assert len(args.moonshine_encoder) == 0, args.moonshine_encoder
assert (
len(args.moonshine_uncached_decoder) == 0
), args.moonshine_uncached_decoder
assert len(args.moonshine_cached_decoder) == 0, args.moonshine_cached_decoder
recognizer = sherpa_onnx.OfflineRecognizer.from_whisper(
encoder=args.whisper_encoder,
@@ -339,6 +410,22 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
task=args.whisper_task,
tail_paddings=args.whisper_tail_paddings,
)
elif args.moonshine_preprocessor:
assert_file_exists(args.moonshine_preprocessor)
assert_file_exists(args.moonshine_encoder)
assert_file_exists(args.moonshine_uncached_decoder)
assert_file_exists(args.moonshine_cached_decoder)
recognizer = sherpa_onnx.OfflineRecognizer.from_moonshine(
preprocessor=args.moonshine_preprocessor,
encoder=args.moonshine_encoder,
uncached_decoder=args.moonshine_uncached_decoder,
cached_decoder=args.moonshine_cached_decoder,
tokens=args.tokens,
num_threads=args.num_threads,
decoding_method=args.decoding_method,
debug=args.debug,
)
else:
raise ValueError("Please specify at least one model")
@@ -424,28 +511,32 @@ def main():
segment_list = []
print("Started!")
start_t = dt.datetime.now()
num_processed_samples = 0
is_silence = False
is_eof = False
# TODO(fangjun): Support multithreads
while True:
# *2 because int16_t has two bytes
data = process.stdout.read(frames_per_read * 2)
if not data:
if is_silence:
if is_eof:
break
is_silence = True
# The converted audio file does not have a mute data of 1 second or more at the end, which will result in the loss of the last segment data
is_eof = True
# pad 1 second at the end of the file for the VAD
data = np.zeros(1 * args.sample_rate, dtype=np.int16)
samples = np.frombuffer(data, dtype=np.int16)
samples = samples.astype(np.float32) / 32768
num_processed_samples += samples.shape[0]
buffer = np.concatenate([buffer, samples])
while len(buffer) > window_size:
vad.accept_waveform(buffer[:window_size])
buffer = buffer[window_size:]
if is_silence:
if is_eof:
vad.flush()
streams = []
@@ -471,6 +562,11 @@ def main():
seg.text = stream.result.text
segment_list.append(seg)
end_t = dt.datetime.now()
elapsed_seconds = (end_t - start_t).total_seconds()
duration = num_processed_samples / 16000
rtf = elapsed_seconds / duration
srt_filename = Path(args.sound_file).with_suffix(".srt")
with open(srt_filename, "w", encoding="utf-8") as f:
for i, seg in enumerate(segment_list):
@@ -479,6 +575,9 @@ def main():
print("", file=f)
print(f"Saved to {srt_filename}")
print(f"Audio duration:\t{duration:.3f} s")
print(f"Elapsed:\t{elapsed_seconds:.3f} s")
print(f"RTF = {elapsed_seconds:.3f}/{duration:.3f} = {rtf:.3f}")
print("Done!")

108
python-api-examples/non_streaming_server.py
View File

@@ -66,7 +66,21 @@ python3 ./python-api-examples/non_streaming_server.py \
--wenet-ctc ./sherpa-onnx-zh-wenet-wenetspeech/model.onnx \
--tokens ./sherpa-onnx-zh-wenet-wenetspeech/tokens.txt
(5) Use a Whisper model
(5) Use a Moonshine model
cd /path/to/sherpa-onnx
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-moonshine-tiny-en-int8.tar.bz2
tar xvf sherpa-onnx-moonshine-tiny-en-int8.tar.bz2
rm sherpa-onnx-moonshine-tiny-en-int8.tar.bz2
python3 ./python-api-examples/non_streaming_server.py \
--moonshine-preprocessor=./sherpa-onnx-moonshine-tiny-en-int8/preprocess.onnx \
--moonshine-encoder=./sherpa-onnx-moonshine-tiny-en-int8/encode.int8.onnx \
--moonshine-uncached-decoder=./sherpa-onnx-moonshine-tiny-en-int8/uncached_decode.int8.onnx \
--moonshine-cached-decoder=./sherpa-onnx-moonshine-tiny-en-int8/cached_decode.int8.onnx \
--tokens=./sherpa-onnx-moonshine-tiny-en-int8/tokens.txt
(6) Use a Whisper model
cd /path/to/sherpa-onnx
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-whisper-tiny.en.tar.bz2
@@ -78,7 +92,7 @@ python3 ./python-api-examples/non_streaming_server.py \
--whisper-decoder=./sherpa-onnx-whisper-tiny.en/tiny.en-decoder.onnx \
--tokens=./sherpa-onnx-whisper-tiny.en/tiny.en-tokens.txt
(5) Use a tdnn model of the yesno recipe from icefall
(7) Use a tdnn model of the yesno recipe from icefall
cd /path/to/sherpa-onnx
@@ -92,7 +106,7 @@ python3 ./python-api-examples/non_streaming_server.py \
--tdnn-model=./sherpa-onnx-tdnn-yesno/model-epoch-14-avg-2.onnx \
--tokens=./sherpa-onnx-tdnn-yesno/tokens.txt
(6) Use a Non-streaming SenseVoice model
(8) Use a Non-streaming SenseVoice model
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17.tar.bz2
tar xvf sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17.tar.bz2
@@ -254,6 +268,36 @@ def add_tdnn_ctc_model_args(parser: argparse.ArgumentParser):
)
def add_moonshine_model_args(parser: argparse.ArgumentParser):
parser.add_argument(
"--moonshine-preprocessor",
default="",
type=str,
help="Path to moonshine preprocessor model",
)
parser.add_argument(
"--moonshine-encoder",
default="",
type=str,
help="Path to moonshine encoder model",
)
parser.add_argument(
"--moonshine-uncached-decoder",
default="",
type=str,
help="Path to moonshine uncached decoder model",
)
parser.add_argument(
"--moonshine-cached-decoder",
default="",
type=str,
help="Path to moonshine cached decoder model",
)
def add_whisper_model_args(parser: argparse.ArgumentParser):
parser.add_argument(
"--whisper-encoder",
@@ -311,6 +355,7 @@ def add_model_args(parser: argparse.ArgumentParser):
add_wenet_ctc_model_args(parser)
add_tdnn_ctc_model_args(parser)
add_whisper_model_args(parser)
add_moonshine_model_args(parser)
parser.add_argument(
"--tokens",
@@ -876,6 +921,12 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
assert len(args.whisper_encoder) == 0, args.whisper_encoder
assert len(args.whisper_decoder) == 0, args.whisper_decoder
assert len(args.tdnn_model) == 0, args.tdnn_model
assert len(args.moonshine_preprocessor) == 0, args.moonshine_preprocessor
assert len(args.moonshine_encoder) == 0, args.moonshine_encoder
assert (
len(args.moonshine_uncached_decoder) == 0
), args.moonshine_uncached_decoder
assert len(args.moonshine_cached_decoder) == 0, args.moonshine_cached_decoder
assert_file_exists(args.encoder)
assert_file_exists(args.decoder)
@@ -903,6 +954,12 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
assert len(args.whisper_encoder) == 0, args.whisper_encoder
assert len(args.whisper_decoder) == 0, args.whisper_decoder
assert len(args.tdnn_model) == 0, args.tdnn_model
assert len(args.moonshine_preprocessor) == 0, args.moonshine_preprocessor
assert len(args.moonshine_encoder) == 0, args.moonshine_encoder
assert (
len(args.moonshine_uncached_decoder) == 0
), args.moonshine_uncached_decoder
assert len(args.moonshine_cached_decoder) == 0, args.moonshine_cached_decoder
assert_file_exists(args.paraformer)
@@ -921,6 +978,12 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
assert len(args.whisper_encoder) == 0, args.whisper_encoder
assert len(args.whisper_decoder) == 0, args.whisper_decoder
assert len(args.tdnn_model) == 0, args.tdnn_model
assert len(args.moonshine_preprocessor) == 0, args.moonshine_preprocessor
assert len(args.moonshine_encoder) == 0, args.moonshine_encoder
assert (
len(args.moonshine_uncached_decoder) == 0
), args.moonshine_uncached_decoder
assert len(args.moonshine_cached_decoder) == 0, args.moonshine_cached_decoder
assert_file_exists(args.sense_voice)
recognizer = sherpa_onnx.OfflineRecognizer.from_sense_voice(
@@ -934,6 +997,12 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
assert len(args.whisper_encoder) == 0, args.whisper_encoder
assert len(args.whisper_decoder) == 0, args.whisper_decoder
assert len(args.tdnn_model) == 0, args.tdnn_model
assert len(args.moonshine_preprocessor) == 0, args.moonshine_preprocessor
assert len(args.moonshine_encoder) == 0, args.moonshine_encoder
assert (
len(args.moonshine_uncached_decoder) == 0
), args.moonshine_uncached_decoder
assert len(args.moonshine_cached_decoder) == 0, args.moonshine_cached_decoder
assert_file_exists(args.nemo_ctc)
@@ -950,6 +1019,12 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
assert len(args.whisper_encoder) == 0, args.whisper_encoder
assert len(args.whisper_decoder) == 0, args.whisper_decoder
assert len(args.tdnn_model) == 0, args.tdnn_model
assert len(args.moonshine_preprocessor) == 0, args.moonshine_preprocessor
assert len(args.moonshine_encoder) == 0, args.moonshine_encoder
assert (
len(args.moonshine_uncached_decoder) == 0
), args.moonshine_uncached_decoder
assert len(args.moonshine_cached_decoder) == 0, args.moonshine_cached_decoder
assert_file_exists(args.wenet_ctc)
@@ -966,6 +1041,12 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
assert len(args.tdnn_model) == 0, args.tdnn_model
assert_file_exists(args.whisper_encoder)
assert_file_exists(args.whisper_decoder)
assert len(args.moonshine_preprocessor) == 0, args.moonshine_preprocessor
assert len(args.moonshine_encoder) == 0, args.moonshine_encoder
assert (
len(args.moonshine_uncached_decoder) == 0
), args.moonshine_uncached_decoder
assert len(args.moonshine_cached_decoder) == 0, args.moonshine_cached_decoder
recognizer = sherpa_onnx.OfflineRecognizer.from_whisper(
encoder=args.whisper_encoder,
@@ -980,6 +1061,12 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
)
elif args.tdnn_model:
assert_file_exists(args.tdnn_model)
assert len(args.moonshine_preprocessor) == 0, args.moonshine_preprocessor
assert len(args.moonshine_encoder) == 0, args.moonshine_encoder
assert (
len(args.moonshine_uncached_decoder) == 0
), args.moonshine_uncached_decoder
assert len(args.moonshine_cached_decoder) == 0, args.moonshine_cached_decoder
recognizer = sherpa_onnx.OfflineRecognizer.from_tdnn_ctc(
model=args.tdnn_model,
@@ -990,6 +1077,21 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
decoding_method=args.decoding_method,
provider=args.provider,
)
elif args.moonshine_preprocessor:
assert_file_exists(args.moonshine_preprocessor)
assert_file_exists(args.moonshine_encoder)
assert_file_exists(args.moonshine_uncached_decoder)
assert_file_exists(args.moonshine_cached_decoder)
recognizer = sherpa_onnx.OfflineRecognizer.from_moonshine(
preprocessor=args.moonshine_preprocessor,
encoder=args.moonshine_encoder,
uncached_decoder=args.moonshine_uncached_decoder,
cached_decoder=args.moonshine_cached_decoder,
tokens=args.tokens,
num_threads=args.num_threads,
decoding_method=args.decoding_method,
)
else:
raise ValueError("Please specify at least one model")

82
python-api-examples/offline-moonshine-decode-files.py Normal file
View File

@@ -0,0 +1,82 @@
#!/usr/bin/env python3
"""
This file shows how to use a non-streaming Moonshine model from
https://github.com/usefulsensors/moonshine
to decode files.
Please download model files from
https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
For instance,
wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-moonshine-tiny-en-int8.tar.bz2
tar xvf sherpa-onnx-moonshine-tiny-en-int8.tar.bz2
rm sherpa-onnx-moonshine-tiny-en-int8.tar.bz2
"""
import datetime as dt
from pathlib import Path
import sherpa_onnx
import soundfile as sf
def create_recognizer():
preprocessor = "./sherpa-onnx-moonshine-tiny-en-int8/preprocess.onnx"
encoder = "./sherpa-onnx-moonshine-tiny-en-int8/encode.int8.onnx"
uncached_decoder = "./sherpa-onnx-moonshine-tiny-en-int8/uncached_decode.int8.onnx"
cached_decoder = "./sherpa-onnx-moonshine-tiny-en-int8/cached_decode.int8.onnx"
tokens = "./sherpa-onnx-moonshine-tiny-en-int8/tokens.txt"
test_wav = "./sherpa-onnx-moonshine-tiny-en-int8/test_wavs/0.wav"
if not Path(preprocessor).is_file() or not Path(test_wav).is_file():
raise ValueError(
"""Please download model files from
https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
"""
)
return (
sherpa_onnx.OfflineRecognizer.from_moonshine(
preprocessor=preprocessor,
encoder=encoder,
uncached_decoder=uncached_decoder,
cached_decoder=cached_decoder,
tokens=tokens,
debug=True,
),
test_wav,
)
def main():
recognizer, wave_filename = create_recognizer()
audio, sample_rate = sf.read(wave_filename, dtype="float32", always_2d=True)
audio = audio[:, 0] # only use the first channel
# audio is a 1-D float32 numpy array normalized to the range [-1, 1]
# sample_rate does not need to be 16000 Hz
start_t = dt.datetime.now()
stream = recognizer.create_stream()
stream.accept_waveform(sample_rate, audio)
recognizer.decode_stream(stream)
end_t = dt.datetime.now()
elapsed_seconds = (end_t - start_t).total_seconds()
duration = audio.shape[-1] / sample_rate
rtf = elapsed_seconds / duration
print(stream.result)
print(wave_filename)
print("Text:", stream.result.text)
print(f"Audio duration:\t{duration:.3f} s")
print(f"Elapsed:\t{elapsed_seconds:.3f} s")
print(f"RTF = {elapsed_seconds:.3f}/{duration:.3f} = {rtf:.3f}")
if __name__ == "__main__":
main()

77
python-api-examples/offline-whisper-decode-files.py Normal file
View File

@@ -0,0 +1,77 @@
#!/usr/bin/env python3
"""
This file shows how to use a non-streaming whisper model from
https://github.com/openai/whisper
to decode files.
Please download model files from
https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
For instance,
wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-whisper-tiny.en.tar.bz2
tar xvf sherpa-onnx-whisper-tiny.en.tar.bz2
rm sherpa-onnx-whisper-tiny.en.tar.bz2
"""
import datetime as dt
from pathlib import Path
import sherpa_onnx
import soundfile as sf
def create_recognizer():
encoder = "./sherpa-onnx-whisper-tiny.en/tiny.en-encoder.int8.onnx"
decoder = "./sherpa-onnx-whisper-tiny.en/tiny.en-decoder.int8.onnx"
tokens = "./sherpa-onnx-whisper-tiny.en/tiny.en-tokens.txt"
test_wav = "./sherpa-onnx-whisper-tiny.en/test_wavs/0.wav"
if not Path(encoder).is_file() or not Path(test_wav).is_file():
raise ValueError(
"""Please download model files from
https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
"""
)
return (
sherpa_onnx.OfflineRecognizer.from_whisper(
encoder=encoder,
decoder=decoder,
tokens=tokens,
debug=True,
),
test_wav,
)
def main():
recognizer, wave_filename = create_recognizer()
audio, sample_rate = sf.read(wave_filename, dtype="float32", always_2d=True)
audio = audio[:, 0] # only use the first channel
# audio is a 1-D float32 numpy array normalized to the range [-1, 1]
# sample_rate does not need to be 16000 Hz
start_t = dt.datetime.now()
stream = recognizer.create_stream()
stream.accept_waveform(sample_rate, audio)
recognizer.decode_stream(stream)
end_t = dt.datetime.now()
elapsed_seconds = (end_t - start_t).total_seconds()
duration = audio.shape[-1] / sample_rate
rtf = elapsed_seconds / duration
print(stream.result)
print(wave_filename)
print("Text:", stream.result.text)
print(f"Audio duration:\t{duration:.3f} s")
print(f"Elapsed:\t{elapsed_seconds:.3f} s")
print(f"RTF = {elapsed_seconds:.3f}/{duration:.3f} = {rtf:.3f}")
if __name__ == "__main__":
main()

83
python-api-examples/vad-with-non-streaming-asr.py
View File

@@ -35,7 +35,18 @@ Note that you need a non-streaming model for this script.
--sample-rate=16000 \
--feature-dim=80
(3) For Whisper models
(3) For Moonshine models
./python-api-examples/vad-with-non-streaming-asr.py \
--silero-vad-model=/path/to/silero_vad.onnx \
--moonshine-preprocessor=./sherpa-onnx-moonshine-tiny-en-int8/preprocess.onnx \
--moonshine-encoder=./sherpa-onnx-moonshine-tiny-en-int8/encode.int8.onnx \
--moonshine-uncached-decoder=./sherpa-onnx-moonshine-tiny-en-int8/uncached_decode.int8.onnx \
--moonshine-cached-decoder=./sherpa-onnx-moonshine-tiny-en-int8/cached_decode.int8.onnx \
--tokens=./sherpa-onnx-moonshine-tiny-en-int8/tokens.txt \
--num-threads=2
(4) For Whisper models
./python-api-examples/vad-with-non-streaming-asr.py \
--silero-vad-model=/path/to/silero_vad.onnx \
@@ -45,7 +56,7 @@ Note that you need a non-streaming model for this script.
--whisper-task=transcribe \
--num-threads=2
(4) For SenseVoice CTC models
(5) For SenseVoice CTC models
./python-api-examples/vad-with-non-streaming-asr.py \
--silero-vad-model=/path/to/silero_vad.onnx \
@@ -192,6 +203,34 @@ def get_args():
""",
)
parser.add_argument(
"--moonshine-preprocessor",
default="",
type=str,
help="Path to moonshine preprocessor model",
)
parser.add_argument(
"--moonshine-encoder",
default="",
type=str,
help="Path to moonshine encoder model",
)
parser.add_argument(
"--moonshine-uncached-decoder",
default="",
type=str,
help="Path to moonshine uncached decoder model",
)
parser.add_argument(
"--moonshine-cached-decoder",
default="",
type=str,
help="Path to moonshine cached decoder model",
)
parser.add_argument(
"--blank-penalty",
type=float,
@@ -251,6 +290,12 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
assert len(args.sense_voice) == 0, args.sense_voice
assert len(args.whisper_encoder) == 0, args.whisper_encoder
assert len(args.whisper_decoder) == 0, args.whisper_decoder
assert len(args.moonshine_preprocessor) == 0, args.moonshine_preprocessor
assert len(args.moonshine_encoder) == 0, args.moonshine_encoder
assert (
len(args.moonshine_uncached_decoder) == 0
), args.moonshine_uncached_decoder
assert len(args.moonshine_cached_decoder) == 0, args.moonshine_cached_decoder
assert_file_exists(args.encoder)
assert_file_exists(args.decoder)
@@ -272,6 +317,12 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
assert len(args.sense_voice) == 0, args.sense_voice
assert len(args.whisper_encoder) == 0, args.whisper_encoder
assert len(args.whisper_decoder) == 0, args.whisper_decoder
assert len(args.moonshine_preprocessor) == 0, args.moonshine_preprocessor
assert len(args.moonshine_encoder) == 0, args.moonshine_encoder
assert (
len(args.moonshine_uncached_decoder) == 0
), args.moonshine_uncached_decoder
assert len(args.moonshine_cached_decoder) == 0, args.moonshine_cached_decoder
assert_file_exists(args.paraformer)
@@ -287,6 +338,12 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
elif args.sense_voice:
assert len(args.whisper_encoder) == 0, args.whisper_encoder
assert len(args.whisper_decoder) == 0, args.whisper_decoder
assert len(args.moonshine_preprocessor) == 0, args.moonshine_preprocessor
assert len(args.moonshine_encoder) == 0, args.moonshine_encoder
assert (
len(args.moonshine_uncached_decoder) == 0
), args.moonshine_uncached_decoder
assert len(args.moonshine_cached_decoder) == 0, args.moonshine_cached_decoder
assert_file_exists(args.sense_voice)
recognizer = sherpa_onnx.OfflineRecognizer.from_sense_voice(
@@ -299,6 +356,12 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
elif args.whisper_encoder:
assert_file_exists(args.whisper_encoder)
assert_file_exists(args.whisper_decoder)
assert len(args.moonshine_preprocessor) == 0, args.moonshine_preprocessor
assert len(args.moonshine_encoder) == 0, args.moonshine_encoder
assert (
len(args.moonshine_uncached_decoder) == 0
), args.moonshine_uncached_decoder
assert len(args.moonshine_cached_decoder) == 0, args.moonshine_cached_decoder
recognizer = sherpa_onnx.OfflineRecognizer.from_whisper(
encoder=args.whisper_encoder,
@@ -311,6 +374,22 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
task=args.whisper_task,
tail_paddings=args.whisper_tail_paddings,
)
elif args.moonshine_preprocessor:
assert_file_exists(args.moonshine_preprocessor)
assert_file_exists(args.moonshine_encoder)
assert_file_exists(args.moonshine_uncached_decoder)
assert_file_exists(args.moonshine_cached_decoder)
recognizer = sherpa_onnx.OfflineRecognizer.from_moonshine(
preprocessor=args.moonshine_preprocessor,
encoder=args.moonshine_encoder,
uncached_decoder=args.moonshine_uncached_decoder,
cached_decoder=args.moonshine_cached_decoder,
tokens=args.tokens,
num_threads=args.num_threads,
decoding_method=args.decoding_method,
debug=args.debug,
)
else:
raise ValueError("Please specify at least one model")