Apply Noise Reduction to Audio for Streaming Speech-to-Text

1
import logging
2
import numpy as np
3
import noisereduce as nr
4
import assemblyai as aai
5
from typing import Type
6
from assemblyai.streaming.v3 import (
7
    BeginEvent,
8
    StreamingClient,
9
    StreamingClientOptions,
10
    StreamingError,
11
    StreamingEvents,
12
    StreamingParameters,
13
    StreamingSessionParameters,
14
    TerminationEvent,
15
    TurnEvent,
16
)
17
18
logging.basicConfig(level=logging.INFO)
19
20
api_key = "<YOUR_API_KEY>"
21
22
# --- Noise-reduced microphone stream ---
23
def noise_reduced_mic_stream(sample_rate=16000):
24
    mic = aai.extras.MicrophoneStream(sample_rate=sample_rate)
25
    buffer = np.array([], dtype=np.int16)
26
    buffer_size = int(sample_rate * 0.5)  # 0.5 seconds
27
28
    for raw_audio in mic:
29
        audio_data = np.frombuffer(raw_audio, dtype=np.int16)
30
        buffer = np.append(buffer, audio_data)
31
32
        if len(buffer) >= buffer_size:
33
            float_audio = buffer.astype(np.float32) / 32768.0
34
            denoised = nr.reduce_noise(
35
                y=float_audio,
36
                sr=sample_rate,
37
                prop_decrease=0.75,
38
                n_fft=1024,
39
            )
40
            int_audio = (denoised * 32768.0).astype(np.int16)
41
            buffer = buffer[-1024:]  # keep some overlap
42
            yield int_audio.tobytes()
43
44
45
# --- Event Handlers ---
46
def on_begin(self: Type[StreamingClient], event: BeginEvent):
47
    print(f" Session started: {event.id}")
48
49
def on_turn(self: Type[StreamingClient], event: TurnEvent):
50
    print(f"{event.transcript} ({event.end_of_turn})")
51
52
    if event.end_of_turn and not event.turn_is_formatted:
53
        self.set_params(StreamingSessionParameters(format_turns=True))
54
55
def on_terminated(self: Type[StreamingClient], event: TerminationEvent):
56
    print(f" Session terminated after {event.audio_duration_seconds} seconds")
57
58
def on_error(self: Type[StreamingClient], error: StreamingError):
59
    print(f" Error occurred: {error}")
60
61
# --- Main Function ---
62
def main():
63
    client = StreamingClient(
64
        StreamingClientOptions(
65
            api_key=api_key,
66
            api_host="streaming.assemblyai.com",
67
        )
68
    )
69
70
    client.on(StreamingEvents.Begin, on_begin)
71
    client.on(StreamingEvents.Turn, on_turn)
72
    client.on(StreamingEvents.Termination, on_terminated)
73
    client.on(StreamingEvents.Error, on_error)
74
75
    client.connect(
76
        StreamingParameters(
77
            sample_rate=16000,
78
            format_turns=True,
79
        )
80
    )
81
82
    try:
83
        denoised_stream = noise_reduced_mic_stream(sample_rate=16000)
84
        client.stream(denoised_stream)
85
    finally:
86
        client.disconnect(terminate=True)
87
88
if __name__ == "__main__":
89
    main()

1
import logging
2
import numpy as np
3
import noisereduce as nr
4
import assemblyai as aai
5
from typing import Type
6
from assemblyai.streaming.v3 import (
7
    BeginEvent,
8
    StreamingClient,
9
    StreamingClientOptions,
10
    StreamingError,
11
    StreamingEvents,
12
    StreamingParameters,
13
    StreamingSessionParameters,
14
    TerminationEvent,
15
    TurnEvent,
16
)

1
api_key = "<YOUR_API_KEY>"

1
def noise_reduced_mic_stream(sample_rate=16000):
2
    mic = aai.extras.MicrophoneStream(sample_rate=sample_rate)
3
    buffer = np.array([], dtype=np.int16)
4
    buffer_size = int(sample_rate * 0.5)  # 0.5 seconds
5
    for raw_audio in mic:
6
        audio_data = np.frombuffer(raw_audio, dtype=np.int16)
7
        buffer = np.append(buffer, audio_data)
8
        if len(buffer) >= buffer_size:
9
            float_audio = buffer.astype(np.float32) / 32768.0
10
            denoised = nr.reduce_noise(
11
                y=float_audio,
12
                sr=sample_rate,
13
                prop_decrease=0.75,
14
                n_fft=1024,
15
            )
16
            int_audio = (denoised * 32768.0).astype(np.int16)
17
            buffer = buffer[-1024:]  # keep some overlap
18
            yield int_audio.tobytes()

1
def on_begin(self: Type[StreamingClient], event: BeginEvent):
2
    print(f" Session started: {event.id}")
3
4
def on_turn(self: Type[StreamingClient], event: TurnEvent):
5
    print(f"{event.transcript} ({event.end_of_turn})")
6
7
    if event.end_of_turn and not event.turn_is_formatted:
8
        self.set_params(StreamingSessionParameters(format_turns=True))
9
10
def on_terminated(self: Type[StreamingClient], event: TerminationEvent):
11
    print(f" Session terminated after {event.audio_duration_seconds} seconds")
12
    
13
def on_error(self: Type[StreamingClient], error: StreamingError):
14
    print(f" Error occurred: {error}")

1
def main():
2
    client = StreamingClient(
3
        StreamingClientOptions(
4
            api_key=api_key,
5
            api_host="streaming.assemblyai.com",
6
        )
7
    )
8
9
    client.on(StreamingEvents.Begin, on_begin)
10
    client.on(StreamingEvents.Turn, on_turn)
11
    client.on(StreamingEvents.Termination, on_terminated)
12
    client.on(StreamingEvents.Error, on_error)
13
14
    client.connect(
15
        StreamingParameters(
16
            sample_rate=16000,
17
            format_turns=True,
18
        )
19
    )
20
21
    try:
22
        denoised_stream = noise_reduced_mic_stream(sample_rate=16000)
23
        client.stream(denoised_stream)
24
    finally:
25
        client.disconnect(terminate=True)
26
27
if __name__ == "__main__":
28
    main()