File size: 7,299 Bytes
ad2cb5b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
import os
import json
import ffmpeg
import whisper
import subprocess
import base64
from pathlib import Path
from typing import Dict, List
import google.generativeai as genai

from config import make_path, GEMINI_API_KEY
from files.utils.logging import get_logger


class AudioAnalyzer:
    def __init__(self, video_path: str, gemini_api_key: str = "", model_size: str = 'small'):
        self.model_size = model_size
        self.video_path = Path(video_path)
        self.audio_path = make_path('interim/audio', video_path, 'audio', 'wav')
        self.json_out = make_path('processed/audio-analysis', video_path, 'audio_analysis', 'json')
        self.logger = get_logger('audio_analysis', f'{self.video_path.stem}_log.txt')

        # ✅ Set Gemini key (explicit or from environment)
        if gemini_api_key:
            genai.configure(api_key=gemini_api_key)
        else:
            genai.configure(api_key=os.getenv("GEMINI_API_KEY", ""))
        self.llm_model = genai.GenerativeModel('gemini-2.5-pro')

    def _extract_audio(self) -> None:
        self.audio_path.parent.mkdir(parents=True, exist_ok=True)
        (
            ffmpeg
            .input(str(self.video_path))
            .output(str(self.audio_path), ac=1, ar='16k', format='wav', loglevel='quiet')
            .overwrite_output()
            .run()
        )
        self.logger.info('Audio extracted to %s', self.audio_path)

    def _transcribe(self) -> Dict:
        model = whisper.load_model(self.model_size)
        return model.transcribe(str(self.audio_path), fp16=False)

    def _loudness_stats(self, audio_path: Path) -> Dict:
        cmd = [
            'ffmpeg', '-i', str(audio_path),
            '-af', 'volumedetect',
            '-f', 'null', 'NUL' if os.name == 'nt' else '/dev/null'
        ]
        result = subprocess.run(cmd, capture_output=True, text=True)
        mean = peak = None
        for line in result.stderr.splitlines():
            if 'mean_volume:' in line:
                mean = float(line.split('mean_volume:')[1].split()[0])
            if 'max_volume:' in line:
                peak = float(line.split('max_volume:')[1].split()[0])
        return {'loudness_mean': mean, 'loudness_peak': peak}

    def _load_visual_context(self) -> Dict:
        """Load nearby frames and brightness values from extracted frame data."""
        frame_json_path = make_path('processed/scene-detection', self.video_path, 'scene', 'json')
        frames_dir = make_path('interim/frames', self.video_path, '', '')

        if not frame_json_path.exists():
            self.logger.warning("Frame metadata not found: %s", frame_json_path)
            return {}

        with open(frame_json_path, 'r', encoding='utf-8') as f:
            scene_data = json.load(f)

        if not scene_data.get('scenes'):
            return {}

        scene = scene_data['scenes'][0]
        mid_time = (float(scene['start_time']) + float(scene['end_time'])) / 2
        scene_idx = 0

        def get_frame_path(tag):
            return frames_dir / f"{self.video_path.stem}_scene_{scene_idx:02}{tag}.jpg"

        def encode_image(p: Path) -> str:
            if p.exists():
                with open(p, 'rb') as f:
                    return base64.b64encode(f.read()).decode('utf-8')
            return ""

        return {
            'mid_time': mid_time,
            'frame': encode_image(get_frame_path('')),
            'prev': encode_image(get_frame_path('_prev')),
            'next': encode_image(get_frame_path('_next')),
            'brightness': float(scene.get('brightness', -1.0))
        }

    def _gemini_audio_analysis(self, text: str, loudness: Dict, wps: float, visuals: Dict) -> Dict:
        """LLM-enhanced audio analysis using audio + first scene frames + metadata"""
        prompt = f"""
        You are an expert video analyst. Based on the transcript, loudness, speaking pace,
        and the first scene's frames (prev, current, next), analyze the audio tone.

        Answer in JSON only:
        {{
        "tone": "calm|excited|angry|funny|sad|neutral",
        "emotion": "joy|sadness|anger|surprise|neutral|mixed",
        "pace": "fast|medium|slow",
        "delivery_score": 0-100,
        "is_hooking_start": true|false,
        "comment": "brief summary of audio performance",
        "is_dark_artistic": true|false,
        "brightness": 0-100
        }}

        Transcript: {text}
        Loudness: {json.dumps(loudness)}
        Words/sec: {wps}
        Frame brightness: {visuals.get('brightness')}
        """

        # ✅ Properly formatted parts for Gemini multimodal prompt
        parts = [{"text": prompt}]
        for tag in ['prev', 'frame', 'next']:
            img_b64 = visuals.get(tag)
            if img_b64:
                parts.append({
                    "inline_data": {
                        "mime_type": "image/jpeg",
                        "data": base64.b64decode(img_b64),
                    }
                })

        try:
            response = self.llm_model.generate_content(
                contents=[{"role": "user", "parts": parts}],
                generation_config={'temperature': 0.3}
            )
            text = getattr(response, 'text', '').strip()
            cleaned = text.replace('```json', '').replace('```', '')
            return json.loads(cleaned)
        except Exception as e:
            error_msg = str(e)
            self.logger.error("LLM call failed: %s", e)
            
            # Check if it's an API key error - if so, raise it to stop the pipeline
            if any(keyword in error_msg.lower() for keyword in ["api_key", "invalid", "401", "403", "authentication", "unauthorized"]):
                raise ValueError(f"Invalid Gemini API key: {error_msg}") from e
            
            # For other errors, return defaults but log the issue
            return {
                "tone": "neutral",
                "emotion": "neutral",
                "pace": "medium",
                "delivery_score": 50,
                "is_hooking_start": False,
                "comment": "LLM analysis failed, using defaults",
                "is_dark_artistic": False,
                "brightness": visuals.get("brightness", -1.0)
            }

    def analyze(self) -> Dict:
        self._extract_audio()
        whisper_res = self._transcribe()
        full_text = whisper_res['text']
        duration_s = whisper_res['segments'][-1]['end'] if whisper_res['segments'] else 0
        wps = round(len(full_text.split()) / duration_s, 2) if duration_s else 0

        loudness = self._loudness_stats(self.audio_path)
        visual_context = self._load_visual_context()
        gemini_analysis = self._gemini_audio_analysis(full_text, loudness, wps, visual_context)

        result = {
            'full_transcript': full_text,
            'duration_seconds': duration_s,
            'word_count': len(full_text.split()),
            'words_per_second': wps,
            **loudness,
            **gemini_analysis
        }

        self.json_out.parent.mkdir(parents=True, exist_ok=True)
        with open(self.json_out, 'w', encoding='utf-8') as f:
            json.dump(result, f, indent=2)

        self.logger.info('Audio + Visual LLM analysis saved to %s', self.json_out)
        return result