Code Examples

1
import os
2
import requests
3
from pydub import AudioSegment
4
5
API_KEY = os.environ["SMALLEST_API_KEY"]
6
ENDPOINT = "https://api.smallest.ai/waves/v1/stt/"
7
8
9
def preprocess_audio(input_path: str, output_path: str) -> str:
10
    audio = AudioSegment.from_file(input_path)
11
    audio = audio.set_frame_rate(16000).set_channels(1).normalize()
12
    audio = audio.strip_silence(silence_len=100, silence_thresh=-40)
13
    audio.export(output_path, format="wav")
14
    return output_path
15
16
17
def transcribe_with_features(audio_path: str) -> dict:
18
    """Transcribe with Pulse + gender, emotion, diarization, utterances."""
19
    with open(audio_path, "rb") as f:
20
        audio_bytes = f.read()
21
    response = requests.post(
22
        ENDPOINT,
23
        params={
24
            "model": "pulse",
25
            "language": "en",
26
            "word_timestamps": "true",
27
            "gender_detection": "true",
28
            "emotion_detection": "true",
29
            "diarize": "true",
30
        },
31
        headers={
32
            "Authorization": f"Bearer {API_KEY}",
33
            "Content-Type": "application/octet-stream",
34
        },
35
        data=audio_bytes,
36
        timeout=120,
37
    )
38
    response.raise_for_status()
39
    return response.json()
40
41
42
def report(result: dict) -> None:
43
    print("=" * 60)
44
    print("TRANSCRIPTION RESULTS")
45
    print("=" * 60)
46
47
    print(f"\nTranscription: {result.get('transcription', 'N/A')}")
48
49
    if result.get("gender") is not None:
50
        print(f"\nGender: {result['gender']}")
51
52
    if result.get("emotions"):
53
        print("\nEmotion scores:")
54
        for emotion, score in result["emotions"].items():
55
            print(f"  {emotion.capitalize()}: {score:.4f}")
56
57
    utterances = result.get("utterances") or []
58
    if utterances:
59
        print(f"\nUtterances ({len(utterances)}):")
60
        for i, u in enumerate(utterances, 1):
61
            print(
62
                f"  [{i}] {u.get('start', 0):.2f}s – {u.get('end', 0):.2f}s "
63
                f"(speaker: {u.get('speaker', 'unknown')})"
64
            )
65
            print(f"      {u.get('text', '')}")
66
67
    words = result.get("words") or []
68
    if words:
69
        print(f"\nWord-level timestamps: {len(words)} words")
70
71
72
if __name__ == "__main__":
73
    raw_path = "input_audio.mp3"
74
    wav_path = "preprocessed.wav"
75
    try:
76
        preprocess_audio(raw_path, wav_path)
77
        result = transcribe_with_features(wav_path)
78
        report(result)
79
    finally:
80
        if os.path.exists(wav_path):
81
            os.remove(wav_path)