Code Examples

1
import os
2
from pydub import AudioSegment
3
from smallestai.waves import WavesClient
4
5
client = WavesClient(api_key=os.getenv("SMALLEST_API_KEY"))
6
7
def preprocess_audio(input_path, output_path):
8
    """
9
    Preprocess audio file to optimal format for Pulse STT:
10
    - Convert to 16 kHz mono WAV
11
    - Normalize audio levels
12
    - Remove leading/trailing silence
13
    """
14
    audio = AudioSegment.from_file(input_path)
15
    audio = audio.set_frame_rate(16000).set_channels(1)
16
    audio = audio.normalize()
17
    audio = audio.strip_silence(silence_len=100, silence_thresh=-40)
18
    audio.export(output_path, format="wav")
19
    print(f"Preprocessed audio saved to: {output_path}")
20
    return output_path
21
22
def transcribe_with_features(audio_path):
23
    """
24
    Transcribe audio with age detection, emotion detection, and utterances.
25
    """
26
    response = client.transcribe(
27
        file_path=audio_path,
28
        model="pulse",
29
        language="en",
30
        word_timestamps=True,
31
        age_detection=True,
32
        gender_detection=True,
33
        emotion_detection=True,
34
        diarize=True
35
    )
36
    
37
    return response
38
39
def process_results(response):
40
    """
41
    Extract and display transcription results.
42
    """
43
    print("=" * 60)
44
    print("TRANSCRIPTION RESULTS")
45
    print("=" * 60)
46
    
47
    print(f"\nTranscription: {response.get('transcription', 'N/A')}")
48
    
49
    if 'age' in response:
50
        print(f"\nAge: {response['age']}")
51
    if 'gender' in response:
52
        print(f"Gender: {response['gender']}")
53
    
54
    if 'emotions' in response:
55
        print("\nEmotion Scores:")
56
        emotions = response['emotions']
57
        for emotion, score in emotions.items():
58
            print(f"  {emotion.capitalize()}: {score:.2f}")
59
    
60
    if 'utterances' in response:
61
        print("\nUtterances (Sentence-level timestamps):")
62
        for i, utterance in enumerate(response['utterances'], 1):
63
            speaker = utterance.get('speaker', 'unknown')
64
            start = utterance.get('start', 0)
65
            end = utterance.get('end', 0)
66
            text = utterance.get('text', '')
67
            print(f"\n  [{i}] Speaker: {speaker}")
68
            print(f"      Time: {start:.2f}s - {end:.2f}s")
69
            print(f"      Text: {text}")
70
    
71
    if 'words' in response:
72
        print(f"\nWord-level timestamps: {len(response['words'])} words")
73
74
if __name__ == "__main__":
75
    input_audio = "input_audio.mp3"
76
    preprocessed_audio = "preprocessed_audio.wav"
77
    
78
    try:
79
        print("Preprocessing audio...")
80
        preprocess_audio(input_audio, preprocessed_audio)
81
        
82
        print("\nTranscribing audio with age, emotion, and utterance detection...")
83
        result = transcribe_with_features(preprocessed_audio)
84
        
85
        process_results(result)
86
        
87
        if os.path.exists(preprocessed_audio):
88
            os.remove(preprocessed_audio)
89
            print("\nCleaned up temporary preprocessed file.")
90
            
91
    except FileNotFoundError:
92
        print(f"Error: Audio file '{input_audio}' not found.")
93
    except Exception as e:
94
        print(f"Error: {str(e)}")