cjones/WhisperLive
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

add: tensorrt backend

Browse Source
This commit is contained in:
makaveli10
2024-01-09 18:09:50 +00:00
parent cafcb04fbc
commit fd86340f30
4 changed files with 1266 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
whisper_live/server_tensorrt.py
+446
View File
@@ -0,0 +1,446 @@
import websockets
import time
import threading
import json
import textwrap
import logging
logging.basicConfig(level = logging.INFO)
from websockets.sync.server import serve
import torch
import numpy as np
import queue
from whisper_live.vad import VoiceActivityDetection
from scipy.io.wavfile import write
import functools
from whisper_live.vad import VoiceActivityDetection
try:
from whisper_live.transcriber_tensorrt import WhisperTRTLLM
except Exception as e:
logging.error("cannot import WhisperTRTLLM")
pass
class TranscriptionServerTRT:
"""
Represents a transcription server that handles incoming audio from clients.
Attributes:
RATE (int): The audio sampling rate (constant) set to 16000.
vad_model (torch.Module): The voice activity detection model.
vad_threshold (float): The voice activity detection threshold.
clients (dict): A dictionary to store connected clients.
websockets (dict): A dictionary to store WebSocket connections.
clients_start_time (dict): A dictionary to track client start times.
max_clients (int): Maximum allowed connected clients.
max_connection_time (int): Maximum allowed connection time in seconds.
"""
RATE = 16000
def __init__(self):
# voice activity detection model
self.clients = {}
self.websockets = {}
self.clients_start_time = {}
self.max_clients = 4
self.max_connection_time = 600
def get_wait_time(self):
"""
Calculate and return the estimated wait time for clients.
Returns:
float: The estimated wait time in minutes.
"""
wait_time = None
for k, v in self.clients_start_time.items():
current_client_time_remaining = self.max_connection_time - (time.time() - v)
if wait_time is None or current_client_time_remaining < wait_time:
wait_time = current_client_time_remaining
return wait_time / 60
def recv_audio(self, websocket, whisper_tensorrt_path=None):
"""
Receive audio chunks from a client in an infinite loop.
Continuously receives audio frames from a connected client
over a WebSocket connection. It processes the audio frames using a
voice activity detection (VAD) model to determine if they contain speech
or not. If the audio frame contains speech, it is added to the client's
audio data for ASR.
If the maximum number of clients is reached, the method sends a
"WAIT" status to the client, indicating that they should wait
until a slot is available.
If a client's connection exceeds the maximum allowed time, it will
be disconnected, and the client's resources will be cleaned up.
Args:
websocket (WebSocket): The WebSocket connection for the client.
Raises:
Exception: If there is an error during the audio frame processing.
"""
self.vad_model = VoiceActivityDetection()
self.vad_threshold = 0.5
logging.info("New client connected")
options = websocket.recv()
options = json.loads(options)
if len(self.clients) >= self.max_clients:
logging.warning("Client Queue Full. Asking client to wait ...")
wait_time = self.get_wait_time()
response = {
"uid": options["uid"],
"status": "WAIT",
"message": wait_time,
}
websocket.send(json.dumps(response))
websocket.close()
del websocket
return
try:
import tensorrt as trt
import tensorrt_llm
except Exception as e:
websocket.send(
json.dumps(
{
"uid": self.client_uid,
"status": "ERROR",
"message": f"TensorRT-LLM not supported on Server yet. Available backends: 'faster_whisper'"
}
)
)
websocket.close()
del websocket
return
client = ServeClient(
websocket,
multilingual=options["multilingual"],
language=options["language"],
task=options["task"],
client_uid=options["uid"],
model_path=whisper_tensorrt_path
)
self.clients[websocket] = client
self.clients_start_time[websocket] = time.time()
no_voice_activity_chunks = 0
while True:
try:
frame_data = websocket.recv()
frame_np = np.frombuffer(frame_data, dtype=np.float32)
# VAD
try:
speech_prob = self.vad_model(torch.from_numpy(frame_np.copy()), self.RATE).item()
if speech_prob < self.vad_threshold:
no_voice_activity_chunks += 1
if no_voice_activity_chunks > 3:
if not self.clients[websocket].eos:
self.clients[websocket].set_eos(True)
time.sleep(0.1) # EOS stop receiving frames for a 100ms(to send output to LLM.)
continue
no_voice_activity_chunks = 0
self.clients[websocket].set_eos(False)
except Exception as e:
logging.error(e)
return
self.clients[websocket].add_frames(frame_np)
elapsed_time = time.time() - self.clients_start_time[websocket]
if elapsed_time >= self.max_connection_time:
self.clients[websocket].disconnect()
logging.warning(f"{self.clients[websocket]} Client disconnected due to overtime.")
self.clients[websocket].cleanup()
self.clients.pop(websocket)
self.clients_start_time.pop(websocket)
websocket.close()
del websocket
break
except Exception as e:
logging.error(e)
self.clients[websocket].cleanup()
self.clients.pop(websocket)
self.clients_start_time.pop(websocket)
logging.info("Connection Closed.")
logging.info(self.clients)
del websocket
break
def run(self, host, port=9090, whisper_tensorrt_path=None):
"""
Run the transcription server.
Args:
host (str): The host address to bind the server.
port (int): The port number to bind the server.
"""
with serve(
functools.partial(
self.recv_audio,
whisper_tensorrt_path=whisper_tensorrt_path
),
host,
port
) as server:
server.serve_forever()
class ServeClient:
"""
Attributes:
RATE (int): The audio sampling rate (constant) set to 16000.
SERVER_READY (str): A constant message indicating that the server is ready.
DISCONNECT (str): A constant message indicating that the client should disconnect.
client_uid (str): A unique identifier for the client.
data (bytes): Accumulated audio data.
frames (bytes): Accumulated audio frames.
language (str): The language for transcription.
task (str): The task type, e.g., "transcribe."
transcriber (WhisperModel): The Whisper model for speech-to-text.
timestamp_offset (float): The offset in audio timestamps.
frames_np (numpy.ndarray): NumPy array to store audio frames.
frames_offset (float): The offset in audio frames.
text (list): List of transcribed text segments.
current_out (str): The current incomplete transcription.
prev_out (str): The previous incomplete transcription.
t_start (float): Timestamp for the start of transcription.
exit (bool): A flag to exit the transcription thread.
same_output_threshold (int): Threshold for consecutive same output segments.
show_prev_out_thresh (int): Threshold for showing previous output segments.
add_pause_thresh (int): Threshold for adding a pause (blank) segment.
transcript (list): List of transcribed segments.
send_last_n_segments (int): Number of last segments to send to the client.
wrapper (textwrap.TextWrapper): Text wrapper for formatting text.
pick_previous_segments (int): Number of previous segments to include in the output.
websocket: The WebSocket connection for the client.
"""
RATE = 16000
SERVER_READY = "SERVER_READY"
DISCONNECT = "DISCONNECT"
def __init__(
self,
websocket,
task="transcribe",
device=None,
multilingual=False,
language=None,
client_uid=None,
model_path=None
):
"""
Initialize a ServeClient instance.
The Whisper model is initialized based on the client's language and device availability.
The transcription thread is started upon initialization. A "SERVER_READY" message is sent
to the client to indicate that the server is ready.
Args:
websocket (WebSocket): The WebSocket connection for the client.
task (str, optional): The task type, e.g., "transcribe." Defaults to "transcribe".
device (str, optional): The device type for Whisper, "cuda" or "cpu". Defaults to None.
multilingual (bool, optional): Whether the client supports multilingual transcription. Defaults to False.
language (str, optional): The language for transcription. Defaults to None.
client_uid (str, optional): A unique identifier for the client. Defaults to None.
"""
self.client_uid = client_uid
self.data = b""
self.frames = b""
self.language = language if multilingual else "en"
self.task = task
self.transcriber = WhisperTRTLLM(model_path, False, "assets", device="cuda")
self.timestamp_offset = 0.0
self.frames_np = None
self.frames_offset = 0.0
self.text = []
self.current_out = ''
self.prev_out = ''
self.t_start=None
self.exit = False
self.same_output_threshold = 0
self.show_prev_out_thresh = 5 # if pause(no output from whisper) show previous output for 5 seconds
self.add_pause_thresh = 3 # add a blank to segment list as a pause(no speech) for 3 seconds
self.transcript = []
self.prompt = None
self.send_last_n_segments = 10
# text formatting
self.wrapper = textwrap.TextWrapper(width=50)
self.pick_previous_segments = 2
# threading
self.websocket = websocket
self.lock = threading.Lock()
self.eos = False
self.trans_thread = threading.Thread(target=self.speech_to_text)
self.trans_thread.start()
self.websocket.send(
json.dumps(
{
"uid": self.client_uid,
"message": self.SERVER_READY
}
)
)
def set_eos(self, eos):
self.lock.acquire()
self.eos = eos
self.lock.release()
def add_frames(self, frame_np):
"""
Add audio frames to the ongoing audio stream buffer.
This method is responsible for maintaining the audio stream buffer, allowing the continuous addition
of audio frames as they are received. It also ensures that the buffer does not exceed a specified size
to prevent excessive memory usage.
If the buffer size exceeds a threshold (45 seconds of audio data), it discards the oldest 30 seconds
of audio data to maintain a reasonable buffer size. If the buffer is empty, it initializes it with the provided
audio frame. The audio stream buffer is used for real-time processing of audio data for transcription.
Args:
frame_np (numpy.ndarray): The audio frame data as a NumPy array.
"""
self.lock.acquire()
if self.frames_np is not None and self.frames_np.shape[0] > 45*self.RATE:
self.frames_offset += 30.0
self.frames_np = self.frames_np[int(30*self.RATE):]
if self.frames_np is None:
self.frames_np = frame_np.copy()
else:
self.frames_np = np.concatenate((self.frames_np, frame_np), axis=0)
self.lock.release()
def speech_to_text(self):
"""
Process an audio stream in an infinite loop, continuously transcribing the speech.
This method continuously receives audio frames, performs real-time transcription, and sends
transcribed segments to the client via a WebSocket connection.
If the client's language is not detected, it waits for 30 seconds of audio input to make a language prediction.
It utilizes the Whisper ASR model to transcribe the audio, continuously processing and streaming results. Segments
are sent to the client in real-time, and a history of segments is maintained to provide context.Pauses in speech
(no output from Whisper) are handled by showing the previous output for a set duration. A blank segment is added if
there is no speech for a specified duration to indicate a pause.
Raises:
Exception: If there is an issue with audio processing or WebSocket communication.
"""
while True:
if self.exit:
logging.info("Exiting speech to text thread")
break
if self.frames_np is None:
time.sleep(0.02) # wait for any audio to arrive
continue
# clip audio if the current chunk exceeds 30 seconds, this basically implies that
# no valid segment for the last 30 seconds from whisper
if self.frames_np[int((self.timestamp_offset - self.frames_offset)*self.RATE):].shape[0] > 25 * self.RATE:
duration = self.frames_np.shape[0] / self.RATE
self.timestamp_offset = self.frames_offset + duration - 5
samples_take = max(0, (self.timestamp_offset - self.frames_offset)*self.RATE)
input_bytes = self.frames_np[int(samples_take):].copy()
duration = input_bytes.shape[0] / self.RATE
if duration<0.4:
continue
try:
input_sample = input_bytes.copy()
mel, duration = self.transcriber.log_mel_spectrogram(input_sample)
last_segment = self.transcriber.transcribe(mel)
segments = []
if len(last_segment):
if len(self.transcript) < self.send_last_n_segments:
segments = self.transcript[:].copy()
else:
segments = self.transcript[-self.send_last_n_segments:].copy()
print(self.transcript, len(self.transcript))
if last_segment is not None:
segments.append({"text": last_segment})
try:
self.websocket.send(
json.dumps({
"uid": self.client_uid,
"segments": segments,
})
)
if self.eos:
print("EOS is true: ", self.timestamp_offset, duration)
if not len(self.transcript):
self.transcript.append({"text": last_segment + " "})
elif self.transcript[-1]["text"].strip() != last_segment:
self.transcript.append({"text": last_segment + " "})
self.timestamp_offset += duration
# self.set_eos(False)
logging.info(
f"[INFO:] Processed : {self.timestamp_offset} seconds / {self.frames_np.shape[0] / self.RATE} seconds"
)
except Exception as e:
logging.error(f"[ERROR]: {e}")
except Exception as e:
logging.error(f"[ERROR]: {e}")
def disconnect(self):
"""
Notify the client of disconnection and send a disconnect message.
This method sends a disconnect message to the client via the WebSocket connection to notify them
that the transcription service is disconnecting gracefully.
"""
self.websocket.send(
json.dumps(
{
"uid": self.client_uid,
"message": self.DISCONNECT
}
)
)
def cleanup(self):
"""
Perform cleanup tasks before exiting the transcription service.
This method performs necessary cleanup tasks, including stopping the transcription thread, marking
the exit flag to indicate the transcription thread should exit gracefully, and destroying resources
associated with the transcription process.
"""
logging.info("Cleaning up.")
self.exit = True
self.transcriber.destroy()
whisper_live/tensorrt_utils.py
+365
View File
@@ -0,0 +1,365 @@
# SPDX-FileCopyrightText: Copyright (c) 2022-2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import logging
import os
from collections import defaultdict
from functools import lru_cache
from pathlib import Path
from subprocess import CalledProcessError, run
from typing import Dict, Iterable, List, Optional, TextIO, Tuple, Union
import kaldialign
import numpy as np
import soundfile
import torch
import torch.nn.functional as F
Pathlike = Union[str, Path]
SAMPLE_RATE = 16000
N_FFT = 400
HOP_LENGTH = 160
CHUNK_LENGTH = 30
N_SAMPLES = CHUNK_LENGTH * SAMPLE_RATE # 480000 samples in a 30-second chunk
def load_audio(file: str, sr: int = SAMPLE_RATE):
"""
Open an audio file and read as mono waveform, resampling as necessary
Parameters
----------
file: str
The audio file to open
sr: int
The sample rate to resample the audio if necessary
Returns
-------
A NumPy array containing the audio waveform, in float32 dtype.
"""
# This launches a subprocess to decode audio while down-mixing
# and resampling as necessary. Requires the ffmpeg CLI in PATH.
# fmt: off
cmd = [
"ffmpeg", "-nostdin", "-threads", "0", "-i", file, "-f", "s16le", "-ac",
"1", "-acodec", "pcm_s16le", "-ar",
str(sr), "-"
]
# fmt: on
try:
out = run(cmd, capture_output=True, check=True).stdout
except CalledProcessError as e:
raise RuntimeError(f"Failed to load audio: {e.stderr.decode()}") from e
return np.frombuffer(out, np.int16).flatten().astype(np.float32) / 32768.0
def load_audio_wav_format(wav_path):
# make sure audio in .wav format
assert wav_path.endswith(
'.wav'), f"Only support .wav format, but got {wav_path}"
waveform, sample_rate = soundfile.read(wav_path)
assert sample_rate == 16000, f"Only support 16k sample rate, but got {sample_rate}"
return waveform, sample_rate
def pad_or_trim(array, length: int = N_SAMPLES, *, axis: int = -1):
"""
Pad or trim the audio array to N_SAMPLES, as expected by the encoder.
"""
if torch.is_tensor(array):
if array.shape[axis] > length:
array = array.index_select(dim=axis,
index=torch.arange(length,
device=array.device))
if array.shape[axis] < length:
pad_widths = [(0, 0)] * array.ndim
pad_widths[axis] = (0, length - array.shape[axis])
array = F.pad(array,
[pad for sizes in pad_widths[::-1] for pad in sizes])
else:
if array.shape[axis] > length:
array = array.take(indices=range(length), axis=axis)
if array.shape[axis] < length:
pad_widths = [(0, 0)] * array.ndim
pad_widths[axis] = (0, length - array.shape[axis])
array = np.pad(array, pad_widths)
return array
@lru_cache(maxsize=None)
def mel_filters(device,
n_mels: int,
mel_filters_dir: str = None) -> torch.Tensor:
"""
load the mel filterbank matrix for projecting STFT into a Mel spectrogram.
Allows decoupling librosa dependency; saved using:
np.savez_compressed(
"mel_filters.npz",
mel_80=librosa.filters.mel(sr=16000, n_fft=400, n_mels=80),
)
"""
assert n_mels in {80, 128}, f"Unsupported n_mels: {n_mels}"
if mel_filters_dir is None:
mel_filters_path = os.path.join(os.path.dirname(__file__), "assets",
"mel_filters.npz")
else:
mel_filters_path = os.path.join(mel_filters_dir, "mel_filters.npz")
with np.load(mel_filters_path) as f:
return torch.from_numpy(f[f"mel_{n_mels}"]).to(device)
def log_mel_spectrogram(
audio: Union[str, np.ndarray, torch.Tensor],
n_mels: int,
padding: int = 0,
device: Optional[Union[str, torch.device]] = None,
return_duration: bool = False,
mel_filters_dir: str = None,
):
"""
Compute the log-Mel spectrogram of
Parameters
----------
audio: Union[str, np.ndarray, torch.Tensor], shape = (*)
The path to audio or either a NumPy array or Tensor containing the audio waveform in 16 kHz
n_mels: int
The number of Mel-frequency filters, only 80 and 128 are supported
padding: int
Number of zero samples to pad to the right
device: Optional[Union[str, torch.device]]
If given, the audio tensor is moved to this device before STFT
Returns
-------
torch.Tensor, shape = (80 or 128, n_frames)
A Tensor that contains the Mel spectrogram
"""
if not torch.is_tensor(audio):
if isinstance(audio, str):
if audio.endswith('.wav'):
audio, _ = load_audio_wav_format(audio)
else:
audio = load_audio(audio)
assert isinstance(audio,
np.ndarray), f"Unsupported audio type: {type(audio)}"
duration = audio.shape[-1] / SAMPLE_RATE
audio = pad_or_trim(audio, N_SAMPLES)
audio = audio.astype(np.float32)
audio = torch.from_numpy(audio)
if device is not None:
audio = audio.to(device)
if padding > 0:
audio = F.pad(audio, (0, padding))
window = torch.hann_window(N_FFT).to(audio.device)
stft = torch.stft(audio,
N_FFT,
HOP_LENGTH,
window=window,
return_complex=True)
magnitudes = stft[..., :-1].abs()**2
filters = mel_filters(audio.device, n_mels, mel_filters_dir)
mel_spec = filters @ magnitudes
log_spec = torch.clamp(mel_spec, min=1e-10).log10()
log_spec = torch.maximum(log_spec, log_spec.max() - 8.0)
log_spec = (log_spec + 4.0) / 4.0
if return_duration:
return log_spec, duration
else:
return log_spec
def store_transcripts(filename: Pathlike, texts: Iterable[Tuple[str, str,
str]]) -> None:
"""Save predicted results and reference transcripts to a file.
https://github.com/k2-fsa/icefall/blob/master/icefall/utils.py
Args:
filename:
File to save the results to.
texts:
An iterable of tuples. The first element is the cur_id, the second is
the reference transcript and the third element is the predicted result.
Returns:
Return None.
"""
with open(filename, "w") as f:
for cut_id, ref, hyp in texts:
print(f"{cut_id}:\tref={ref}", file=f)
print(f"{cut_id}:\thyp={hyp}", file=f)
def write_error_stats(
f: TextIO,
test_set_name: str,
results: List[Tuple[str, str]],
enable_log: bool = True,
) -> float:
"""Write statistics based on predicted results and reference transcripts.
https://github.com/k2-fsa/icefall/blob/master/icefall/utils.py
It will write the following to the given file:
- WER
- number of insertions, deletions, substitutions, corrects and total
reference words. For example::
Errors: 23 insertions, 57 deletions, 212 substitutions, over 2606
reference words (2337 correct)
- The difference between the reference transcript and predicted result.
An instance is given below::
THE ASSOCIATION OF (EDISON->ADDISON) ILLUMINATING COMPANIES
The above example shows that the reference word is `EDISON`,
but it is predicted to `ADDISON` (a substitution error).
Another example is::
FOR THE FIRST DAY (SIR->*) I THINK
The reference word `SIR` is missing in the predicted
results (a deletion error).
results:
An iterable of tuples. The first element is the cur_id, the second is
the reference transcript and the third element is the predicted result.
enable_log:
If True, also print detailed WER to the console.
Otherwise, it is written only to the given file.
Returns:
Return None.
"""
subs: Dict[Tuple[str, str], int] = defaultdict(int)
ins: Dict[str, int] = defaultdict(int)
dels: Dict[str, int] = defaultdict(int)
# `words` stores counts per word, as follows:
# corr, ref_sub, hyp_sub, ins, dels
words: Dict[str, List[int]] = defaultdict(lambda: [0, 0, 0, 0, 0])
num_corr = 0
ERR = "*"
for cut_id, ref, hyp in results:
ali = kaldialign.align(ref, hyp, ERR)
for ref_word, hyp_word in ali:
if ref_word == ERR:
ins[hyp_word] += 1
words[hyp_word][3] += 1
elif hyp_word == ERR:
dels[ref_word] += 1
words[ref_word][4] += 1
elif hyp_word != ref_word:
subs[(ref_word, hyp_word)] += 1
words[ref_word][1] += 1
words[hyp_word][2] += 1
else:
words[ref_word][0] += 1
num_corr += 1
ref_len = sum([len(r) for _, r, _ in results])
sub_errs = sum(subs.values())
ins_errs = sum(ins.values())
del_errs = sum(dels.values())
tot_errs = sub_errs + ins_errs + del_errs
tot_err_rate = "%.2f" % (100.0 * tot_errs / ref_len)
if enable_log:
logging.info(f"[{test_set_name}] %WER {tot_errs / ref_len:.2%} "
f"[{tot_errs} / {ref_len}, {ins_errs} ins, "
f"{del_errs} del, {sub_errs} sub ]")
print(f"%WER = {tot_err_rate}", file=f)
print(
f"Errors: {ins_errs} insertions, {del_errs} deletions, "
f"{sub_errs} substitutions, over {ref_len} reference "
f"words ({num_corr} correct)",
file=f,
)
print(
"Search below for sections starting with PER-UTT DETAILS:, "
"SUBSTITUTIONS:, DELETIONS:, INSERTIONS:, PER-WORD STATS:",
file=f,
)
print("", file=f)
print("PER-UTT DETAILS: corr or (ref->hyp) ", file=f)
for cut_id, ref, hyp in results:
ali = kaldialign.align(ref, hyp, ERR)
combine_successive_errors = True
if combine_successive_errors:
ali = [[[x], [y]] for x, y in ali]
for i in range(len(ali) - 1):
if ali[i][0] != ali[i][1] and ali[i + 1][0] != ali[i + 1][1]:
ali[i + 1][0] = ali[i][0] + ali[i + 1][0]
ali[i + 1][1] = ali[i][1] + ali[i + 1][1]
ali[i] = [[], []]
ali = [[
list(filter(lambda a: a != ERR, x)),
list(filter(lambda a: a != ERR, y)),
] for x, y in ali]
ali = list(filter(lambda x: x != [[], []], ali))
ali = [[
ERR if x == [] else " ".join(x),
ERR if y == [] else " ".join(y),
] for x, y in ali]
print(
f"{cut_id}:\t" + " ".join((ref_word if ref_word == hyp_word else
f"({ref_word}->{hyp_word})"
for ref_word, hyp_word in ali)),
file=f,
)
print("", file=f)
print("SUBSTITUTIONS: count ref -> hyp", file=f)
for count, (ref, hyp) in sorted([(v, k) for k, v in subs.items()],
reverse=True):
print(f"{count} {ref} -> {hyp}", file=f)
print("", file=f)
print("DELETIONS: count ref", file=f)
for count, ref in sorted([(v, k) for k, v in dels.items()], reverse=True):
print(f"{count} {ref}", file=f)
print("", file=f)
print("INSERTIONS: count hyp", file=f)
for count, hyp in sorted([(v, k) for k, v in ins.items()], reverse=True):
print(f"{count} {hyp}", file=f)
print("", file=f)
print("PER-WORD STATS: word corr tot_errs count_in_ref count_in_hyp",
file=f)
for _, word, counts in sorted([(sum(v[1:]), k, v)
for k, v in words.items()],
reverse=True):
(corr, ref_sub, hyp_sub, ins, dels) = counts
tot_errs = ref_sub + hyp_sub + ins + dels
ref_count = corr + ref_sub + dels
hyp_count = corr + hyp_sub + ins
print(f"{word} {corr} {tot_errs} {ref_count} {hyp_count}", file=f)
return float(tot_err_rate)
whisper_live/transcriber_tensorrt.py
+337
View File
@@ -0,0 +1,337 @@
import argparse
import json
import re
import time
from collections import OrderedDict
from pathlib import Path
from typing import Dict, Iterable, List, Optional, TextIO, Tuple, Union
import torch
import numpy as np
from whisper.tokenizer import get_tokenizer
from whisper_live.tensorrt_utils import (mel_filters, store_transcripts,
write_error_stats, load_audio_wav_format,
pad_or_trim)
import tensorrt_llm
import tensorrt_llm.logger as logger
from tensorrt_llm._utils import (str_dtype_to_torch, str_dtype_to_trt,
trt_dtype_to_torch)
from tensorrt_llm.runtime import ModelConfig, SamplingConfig
from tensorrt_llm.runtime.session import Session, TensorInfo
SAMPLE_RATE = 16000
N_FFT = 400
HOP_LENGTH = 160
CHUNK_LENGTH = 30
N_SAMPLES = CHUNK_LENGTH * SAMPLE_RATE # 480000 samples in a 30-second chunk
class WhisperEncoding:
def __init__(self, engine_dir):
self.session = self.get_session(engine_dir)
def get_session(self, engine_dir):
config_path = engine_dir / 'encoder_config.json'
with open(config_path, 'r') as f:
config = json.load(f)
use_gpt_attention_plugin = config['plugin_config'][
'gpt_attention_plugin']
dtype = config['builder_config']['precision']
n_mels = config['builder_config']['n_mels']
num_languages = config['builder_config']['num_languages']
self.dtype = dtype
self.n_mels = n_mels
self.num_languages = num_languages
serialize_path = engine_dir / f'whisper_encoder_{self.dtype}_tp1_rank0.engine'
with open(serialize_path, 'rb') as f:
session = Session.from_serialized_engine(f.read())
return session
def get_audio_features(self, mel):
inputs = OrderedDict()
output_list = []
inputs.update({'x': mel})
output_list.append(
TensorInfo('x', str_dtype_to_trt(self.dtype), mel.shape))
output_info = (self.session).infer_shapes(output_list)
logger.debug(f'output info {output_info}')
outputs = {
t.name: torch.empty(tuple(t.shape),
dtype=trt_dtype_to_torch(t.dtype),
device='cuda')
for t in output_info
}
stream = torch.cuda.current_stream()
ok = self.session.run(inputs=inputs,
outputs=outputs,
stream=stream.cuda_stream)
assert ok, 'Engine execution failed'
stream.synchronize()
audio_features = outputs['output']
return audio_features
class WhisperDecoding:
def __init__(self, engine_dir, runtime_mapping, debug_mode=False):
self.decoder_config = self.get_config(engine_dir)
self.decoder_generation_session = self.get_session(
engine_dir, runtime_mapping, debug_mode)
def get_config(self, engine_dir):
config_path = engine_dir / 'decoder_config.json'
with open(config_path, 'r') as f:
config = json.load(f)
decoder_config = OrderedDict()
decoder_config.update(config['plugin_config'])
decoder_config.update(config['builder_config'])
return decoder_config
def get_session(self, engine_dir, runtime_mapping, debug_mode=False):
dtype = self.decoder_config['precision']
serialize_path = engine_dir / f'whisper_decoder_{dtype}_tp1_rank0.engine'
with open(serialize_path, "rb") as f:
decoder_engine_buffer = f.read()
decoder_model_config = ModelConfig(
num_heads=self.decoder_config['num_heads'],
num_kv_heads=self.decoder_config['num_heads'],
hidden_size=self.decoder_config['hidden_size'],
vocab_size=self.decoder_config['vocab_size'],
num_layers=self.decoder_config['num_layers'],
gpt_attention_plugin=self.decoder_config['gpt_attention_plugin'],
remove_input_padding=self.decoder_config['remove_input_padding'],
cross_attention=self.decoder_config['cross_attention'],
has_position_embedding=self.
decoder_config['has_position_embedding'],
has_token_type_embedding=self.
decoder_config['has_token_type_embedding'],
)
decoder_generation_session = tensorrt_llm.runtime.GenerationSession(
decoder_model_config,
decoder_engine_buffer,
runtime_mapping,
debug_mode=debug_mode)
return decoder_generation_session
def generate(self,
decoder_input_ids,
encoder_outputs,
eot_id,
max_new_tokens=40,
num_beams=1):
encoder_input_lengths = torch.tensor(
[encoder_outputs.shape[1] for x in range(encoder_outputs.shape[0])],
dtype=torch.int32,
device='cuda')
decoder_input_lengths = torch.tensor([
decoder_input_ids.shape[-1]
for _ in range(decoder_input_ids.shape[0])
],
dtype=torch.int32,
device='cuda')
decoder_max_input_length = torch.max(decoder_input_lengths).item()
# generation config
sampling_config = SamplingConfig(end_id=eot_id,
pad_id=eot_id,
num_beams=num_beams)
self.decoder_generation_session.setup(
decoder_input_lengths.size(0),
decoder_max_input_length,
max_new_tokens,
beam_width=num_beams,
encoder_max_input_length=encoder_outputs.shape[1])
torch.cuda.synchronize()
decoder_input_ids = decoder_input_ids.type(torch.int32).cuda()
output_ids = self.decoder_generation_session.decode(
decoder_input_ids,
decoder_input_lengths,
sampling_config,
encoder_output=encoder_outputs,
encoder_input_lengths=encoder_input_lengths,
)
torch.cuda.synchronize()
# get the list of int from output_ids tensor
output_ids = output_ids.cpu().numpy().tolist()
return output_ids
class WhisperTRTLLM(object):
def __init__(
self,
engine_dir,
debug_mode=False,
assets_dir=None,
device=None
):
world_size = 1
runtime_rank = tensorrt_llm.mpi_rank()
runtime_mapping = tensorrt_llm.Mapping(world_size, runtime_rank)
torch.cuda.set_device(runtime_rank % runtime_mapping.gpus_per_node)
engine_dir = Path(engine_dir)
self.encoder = WhisperEncoding(engine_dir)
self.decoder = WhisperDecoding(engine_dir,
runtime_mapping,
debug_mode=False)
self.n_mels = self.encoder.n_mels
# self.tokenizer = get_tokenizer(num_languages=self.encoder.num_languages,
# tokenizer_dir=assets_dir)
self.device = device
self.tokenizer = get_tokenizer(
False,
num_languages=self.encoder.num_languages,
language="en",
task="transcribe",
)
self.filters = mel_filters(self.device, self.encoder.n_mels, assets_dir)
def log_mel_spectrogram(
self,
audio: Union[str, np.ndarray, torch.Tensor],
padding: int = 0,
return_duration = True
):
"""
Compute the log-Mel spectrogram of
Parameters
----------
audio: Union[str, np.ndarray, torch.Tensor], shape = (*)
The path to audio or either a NumPy array or Tensor containing the audio waveform in 16 kHz
n_mels: int
The number of Mel-frequency filters, only 80 and 128 are supported
padding: int
Number of zero samples to pad to the right
device: Optional[Union[str, torch.device]]
If given, the audio tensor is moved to this device before STFT
Returns
-------
torch.Tensor, shape = (80 or 128, n_frames)
A Tensor that contains the Mel spectrogram
"""
if not torch.is_tensor(audio):
if isinstance(audio, str):
if audio.endswith('.wav'):
audio, _ = load_audio_wav_format(audio)
else:
audio = load_audio(audio)
assert isinstance(audio,
np.ndarray), f"Unsupported audio type: {type(audio)}"
duration = audio.shape[-1] / SAMPLE_RATE
audio = pad_or_trim(audio, N_SAMPLES)
audio = audio.astype(np.float32)
audio = torch.from_numpy(audio)
if self.device is not None:
audio = audio.to(self.device)
if padding > 0:
audio = F.pad(audio, (0, padding))
window = torch.hann_window(N_FFT).to(audio.device)
stft = torch.stft(audio,
N_FFT,
HOP_LENGTH,
window=window,
return_complex=True)
magnitudes = stft[..., :-1].abs()**2
mel_spec = self.filters @ magnitudes
log_spec = torch.clamp(mel_spec, min=1e-10).log10()
log_spec = torch.maximum(log_spec, log_spec.max() - 8.0)
log_spec = (log_spec + 4.0) / 4.0
if return_duration:
return log_spec, duration
else:
return log_spec
def process_batch(
self,
mel,
text_prefix="<|startoftranscript|><|en|><|transcribe|><|notimestamps|>",
num_beams=1):
prompt_id = self.tokenizer.encode(
text_prefix, allowed_special=set(self.tokenizer.special_tokens.keys()))
prompt_id = torch.tensor(prompt_id)
batch_size = mel.shape[0]
decoder_input_ids = prompt_id.repeat(batch_size, 1)
encoder_output = self.encoder.get_audio_features(mel)
output_ids = self.decoder.generate(decoder_input_ids,
encoder_output,
self.tokenizer.eot,
max_new_tokens=96,
num_beams=num_beams)
texts = []
for i in range(len(output_ids)):
text = self.tokenizer.decode(output_ids[i][0]).strip()
texts.append(text)
return texts
def transcribe(
self,
mel,
text_prefix="<|startoftranscript|><|en|><|transcribe|><|notimestamps|>",
dtype='float16',
batch_size=1,
num_beams=1,
):
mel = mel.type(str_dtype_to_torch(dtype))
mel = mel.unsqueeze(0)
predictions = self.process_batch(mel, text_prefix, num_beams)
prediction = predictions[0]
# remove all special tokens in the prediction
prediction = re.sub(r'<\|.*?\|>', '', prediction)
return prediction.strip()
def decode_wav_file(
model,
mel,
text_prefix="<|startoftranscript|><|en|><|transcribe|><|notimestamps|>",
dtype='float16',
batch_size=1,
num_beams=1,
normalizer=None,
mel_filters_dir=None):
mel = mel.type(str_dtype_to_torch(dtype))
mel = mel.unsqueeze(0)
# repeat the mel spectrogram to match the batch size
mel = mel.repeat(batch_size, 1, 1)
predictions = model.process_batch(mel, text_prefix, num_beams)
prediction = predictions[0]
# remove all special tokens in the prediction
prediction = re.sub(r'<\|.*?\|>', '', prediction)
if normalizer:
prediction = normalizer(prediction)
return prediction.strip()
whisper_live/vad.py
+118
View File
@@ -0,0 +1,118 @@
# original: https://github.com/snakers4/silero-vad/blob/master/utils_vad.py
import os
import subprocess
import torch
import numpy as np
import onnxruntime
class VoiceActivityDetection():
def __init__(self, force_onnx_cpu=True):
print("downloading ONNX model...")
path = self.download()
print("loading session")
opts = onnxruntime.SessionOptions()
opts.log_severity_level = 3
opts.inter_op_num_threads = 1
opts.intra_op_num_threads = 1
print("loading onnx model")
if force_onnx_cpu and 'CPUExecutionProvider' in onnxruntime.get_available_providers():
self.session = onnxruntime.InferenceSession(path, providers=['CPUExecutionProvider'], sess_options=opts)
else:
self.session = onnxruntime.InferenceSession(path, providers=['CUDAExecutionProvider'], sess_options=opts)
print("reset states")
self.reset_states()
self.sample_rates = [8000, 16000]
def _validate_input(self, x, sr: int):
if x.dim() == 1:
x = x.unsqueeze(0)
if x.dim() > 2:
raise ValueError(f"Too many dimensions for input audio chunk {x.dim()}")
if sr != 16000 and (sr % 16000 == 0):
step = sr // 16000
x = x[:,::step]
sr = 16000
if sr not in self.sample_rates:
raise ValueError(f"Supported sampling rates: {self.sample_rates} (or multiply of 16000)")
if sr / x.shape[1] > 31.25:
raise ValueError("Input audio chunk is too short")
return x, sr
def reset_states(self, batch_size=1):
self._h = np.zeros((2, batch_size, 64)).astype('float32')
self._c = np.zeros((2, batch_size, 64)).astype('float32')
self._last_sr = 0
self._last_batch_size = 0
def __call__(self, x, sr: int):
x, sr = self._validate_input(x, sr)
batch_size = x.shape[0]
if not self._last_batch_size:
self.reset_states(batch_size)
if (self._last_sr) and (self._last_sr != sr):
self.reset_states(batch_size)
if (self._last_batch_size) and (self._last_batch_size != batch_size):
self.reset_states(batch_size)
if sr in [8000, 16000]:
ort_inputs = {'input': x.numpy(), 'h': self._h, 'c': self._c, 'sr': np.array(sr, dtype='int64')}
ort_outs = self.session.run(None, ort_inputs)
out, self._h, self._c = ort_outs
else:
raise ValueError()
self._last_sr = sr
self._last_batch_size = batch_size
out = torch.tensor(out)
return out
def audio_forward(self, x, sr: int, num_samples: int = 512):
outs = []
x, sr = self._validate_input(x, sr)
if x.shape[1] % num_samples:
pad_num = num_samples - (x.shape[1] % num_samples)
x = torch.nn.functional.pad(x, (0, pad_num), 'constant', value=0.0)
self.reset_states(x.shape[0])
for i in range(0, x.shape[1], num_samples):
wavs_batch = x[:, i:i+num_samples]
out_chunk = self.__call__(wavs_batch, sr)
outs.append(out_chunk)
stacked = torch.cat(outs, dim=1)
return stacked.cpu()
@staticmethod
def download(model_url="https://github.com/snakers4/silero-vad/raw/master/files/silero_vad.onnx"):
target_dir = os.path.expanduser("~/.cache/whisper-live/")
# Ensure the target directory exists
os.makedirs(target_dir, exist_ok=True)
# Define the target file path
model_filename = os.path.join(target_dir, "silero_vad.onnx")
# Check if the model file already exists
if not os.path.exists(model_filename):
# If it doesn't exist, download the model using wget
print("Downloading VAD ONNX model...")
try:
subprocess.run(["wget", "-O", model_filename, model_url], check=True)
except subprocess.CalledProcessError:
print("Failed to download the model using wget.")
return model_filename