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 gender 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
        gender_detection=True,
32
        emotion_detection=True,
33
        diarize=True
34
    )
35
    
36
    return response
37
38
def process_results(response):
39
    """
40
    Extract and display transcription results.
41
    """
42
    print("=" * 60)
43
    print("TRANSCRIPTION RESULTS")
44
    print("=" * 60)
45
    
46
    print(f"\nTranscription: {response.get('transcription', 'N/A')}")
47
    
48
    if 'gender' in response:
49
        print(f"\nGender: {response['gender']}")
50
    
51
    if 'emotions' in response:
52
        print("\nEmotion Scores:")
53
        emotions = response['emotions']
54
        for emotion, score in emotions.items():
55
            print(f"  {emotion.capitalize()}: {score:.2f}")
56
    
57
    if 'utterances' in response:
58
        print("\nUtterances (Sentence-level timestamps):")
59
        for i, utterance in enumerate(response['utterances'], 1):
60
            speaker = utterance.get('speaker', 'unknown')
61
            start = utterance.get('start', 0)
62
            end = utterance.get('end', 0)
63
            text = utterance.get('text', '')
64
            print(f"\n  [{i}] Speaker: {speaker}")
65
            print(f"      Time: {start:.2f}s - {end:.2f}s")
66
            print(f"      Text: {text}")
67
    
68
    if 'words' in response:
69
        print(f"\nWord-level timestamps: {len(response['words'])} words")
70
71
if __name__ == "__main__":
72
    input_audio = "input_audio.mp3"
73
    preprocessed_audio = "preprocessed_audio.wav"
74
    
75
    try:
76
        print("Preprocessing audio...")
77
        preprocess_audio(input_audio, preprocessed_audio)
78
        
79
        print("\nTranscribing audio with gender, emotion, and utterance detection...")
80
        result = transcribe_with_features(preprocessed_audio)
81
        
82
        process_results(result)
83
        
84
        if os.path.exists(preprocessed_audio):
85
            os.remove(preprocessed_audio)
86
            print("\nCleaned up temporary preprocessed file.")
87
            
88
    except FileNotFoundError:
89
        print(f"Error: Audio file '{input_audio}' not found.")
90
    except Exception as e:
91
        print(f"Error: {str(e)}")