Update app.py
Browse files
app.py
CHANGED
|
@@ -25,15 +25,15 @@ lang_pipe = pipeline(
|
|
| 25 |
def convert_to_wav(audio_path):
|
| 26 |
"""Converte qualquer arquivo de áudio para WAV mono 16 kHz."""
|
| 27 |
try:
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
except Exception as e:
|
| 35 |
-
|
| 36 |
-
|
| 37 |
|
| 38 |
def make_speech_head_wav(input_wav_path, max_seconds=7):
|
| 39 |
"""
|
|
@@ -41,47 +41,47 @@ def make_speech_head_wav(input_wav_path, max_seconds=7):
|
|
| 41 |
après suppression du silence initial (avec protection contre les boucles).
|
| 42 |
"""
|
| 43 |
try:
|
| 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 |
except Exception as e:
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
|
| 85 |
|
| 86 |
def detect_language_on_upload(filepath):
|
| 87 |
"""
|
|
@@ -91,74 +91,74 @@ def detect_language_on_upload(filepath):
|
|
| 91 |
return "auto"
|
| 92 |
|
| 93 |
try:
|
| 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 |
except Exception as e:
|
| 146 |
-
|
| 147 |
-
|
| 148 |
|
| 149 |
def ensure_mp3_same_name_as_input(input_path, source_wav_path):
|
| 150 |
"""
|
| 151 |
Cria um arquivo MP3 com o mesmo nome base do arquivo de entrada.
|
| 152 |
"""
|
| 153 |
try:
|
| 154 |
-
|
| 155 |
-
|
| 156 |
-
|
| 157 |
-
|
| 158 |
-
|
| 159 |
except Exception as e:
|
| 160 |
-
|
| 161 |
-
|
| 162 |
|
| 163 |
# --- Fonction principale ---
|
| 164 |
|
|
@@ -191,47 +191,47 @@ with gr.Blocks() as demo:
|
|
| 191 |
|
| 192 |
gr.Markdown("""
|
| 193 |
## ⚡ **Version optimisée**
|
| 194 |
-
- **Détection rapide** : Analyse les
|
| 195 |
- **Robuste** : Fonctionne avec tous types de fichiers
|
| 196 |
- **Timeout protection** : Évite les blocages
|
| 197 |
""")
|
| 198 |
|
| 199 |
with gr.Row():
|
| 200 |
-
|
| 201 |
-
|
| 202 |
-
|
| 203 |
-
|
| 204 |
-
|
| 205 |
-
|
| 206 |
-
|
| 207 |
-
|
| 208 |
-
|
| 209 |
-
|
| 210 |
-
|
| 211 |
-
|
| 212 |
-
|
| 213 |
-
|
| 214 |
|
| 215 |
# Détection automatique de langue lors de l'upload
|
| 216 |
audio_input.change(
|
| 217 |
-
|
| 218 |
-
|
| 219 |
-
|
| 220 |
)
|
| 221 |
|
| 222 |
submit_btn.click(
|
| 223 |
-
|
| 224 |
-
|
| 225 |
-
|
| 226 |
)
|
| 227 |
|
| 228 |
def reset_fields():
|
| 229 |
-
|
| 230 |
|
| 231 |
reset_btn.click(
|
| 232 |
-
|
| 233 |
-
|
| 234 |
-
|
| 235 |
)
|
| 236 |
|
| 237 |
demo.launch(share=True)
|
|
|
|
| 25 |
def convert_to_wav(audio_path):
|
| 26 |
"""Converte qualquer arquivo de áudio para WAV mono 16 kHz."""
|
| 27 |
try:
|
| 28 |
+
audio = AudioSegment.from_file(audio_path)
|
| 29 |
+
audio = audio.set_channels(1)
|
| 30 |
+
audio = audio.set_frame_rate(16000)
|
| 31 |
+
wav_path = os.path.splitext(audio_path)[0] + ".wav"
|
| 32 |
+
audio.export(wav_path, format="wav")
|
| 33 |
+
return wav_path
|
| 34 |
except Exception as e:
|
| 35 |
+
print(f"Erro ao converter para WAV: {e}")
|
| 36 |
+
return None
|
| 37 |
|
| 38 |
def make_speech_head_wav(input_wav_path, max_seconds=7):
|
| 39 |
"""
|
|
|
|
| 41 |
après suppression du silence initial (avec protection contre les boucles).
|
| 42 |
"""
|
| 43 |
try:
|
| 44 |
+
audio = AudioSegment.from_wav(input_wav_path)
|
| 45 |
+
|
| 46 |
+
# Si l'audio est déjà très court, le retourner tel quel
|
| 47 |
+
if len(audio) <= max_seconds * 1000:
|
| 48 |
+
return input_wav_path
|
| 49 |
+
|
| 50 |
+
# Normalisation simple
|
| 51 |
+
normalized = effects.normalize(audio)
|
| 52 |
+
|
| 53 |
+
# Tentative rapide de suppression du silence initial
|
| 54 |
+
try:
|
| 55 |
+
silence_thresh = normalized.dBFS - 20
|
| 56 |
+
# Limiter la recherche aux 30 premières secondes maximum
|
| 57 |
+
search_audio = normalized[:30000]
|
| 58 |
+
start_trim = silence.detect_leading_silence(
|
| 59 |
+
search_audio,
|
| 60 |
+
silence_threshold=silence_thresh,
|
| 61 |
+
chunk_size=100
|
| 62 |
+
)
|
| 63 |
+
# Limiter le trim à 15 secondes max pour éviter les cas extrêmes
|
| 64 |
+
start_trim = min(start_trim, 15000)
|
| 65 |
+
trimmed = normalized[start_trim:]
|
| 66 |
+
except:
|
| 67 |
+
# En cas d'erreur, utiliser l'audio original
|
| 68 |
+
trimmed = normalized
|
| 69 |
+
|
| 70 |
+
# Si trop court après trim, utiliser l'original
|
| 71 |
+
if len(trimmed) < 2000: # Moins de 2 secondes
|
| 72 |
+
trimmed = normalized
|
| 73 |
+
|
| 74 |
+
# Prendre simplement les premiers max_seconds
|
| 75 |
+
clip = trimmed[:max_seconds * 1000]
|
| 76 |
+
|
| 77 |
+
short_path = os.path.splitext(input_wav_path)[0] + f"_head_{max_seconds}s.wav"
|
| 78 |
+
clip.export(short_path, format="wav")
|
| 79 |
+
return short_path
|
| 80 |
|
| 81 |
except Exception as e:
|
| 82 |
+
print(f"Erro ao criar o trecho: {e}")
|
| 83 |
+
# En cas d'erreur, retourner le fichier original
|
| 84 |
+
return input_wav_path
|
| 85 |
|
| 86 |
def detect_language_on_upload(filepath):
|
| 87 |
"""
|
|
|
|
| 91 |
return "auto"
|
| 92 |
|
| 93 |
try:
|
| 94 |
+
print(f"Début détection langue pour: {filepath}")
|
| 95 |
+
|
| 96 |
+
wav_filepath = convert_to_wav(filepath)
|
| 97 |
+
if not wav_filepath:
|
| 98 |
+
print("Échec conversion WAV")
|
| 99 |
+
return "auto"
|
| 100 |
+
|
| 101 |
+
# Créer un extrait court (5 secondes max)
|
| 102 |
+
short_wav = make_speech_head_wav(wav_filepath, max_seconds=7)
|
| 103 |
+
if not short_wav:
|
| 104 |
+
short_wav = wav_filepath
|
| 105 |
+
|
| 106 |
+
print(f"Analyse du fichier: {short_wav}")
|
| 107 |
+
|
| 108 |
+
# Transcrição avec paramètres conservateurs
|
| 109 |
+
outputs = lang_pipe(
|
| 110 |
+
short_wav,
|
| 111 |
+
chunk_length_s=5,
|
| 112 |
+
return_timestamps=False,
|
| 113 |
+
generate_kwargs={"max_new_tokens": 50} # Limiter pour éviter les timeouts
|
| 114 |
+
)
|
| 115 |
+
|
| 116 |
+
transcribed_text = outputs.get("text", "").strip()
|
| 117 |
+
print(f"Texte transcrit: {transcribed_text[:100]}...")
|
| 118 |
+
|
| 119 |
+
# Priorité au language détecté par Whisper
|
| 120 |
+
whisper_lang = outputs.get("language")
|
| 121 |
+
if whisper_lang and isinstance(whisper_lang, str) and len(whisper_lang) <= 5:
|
| 122 |
+
print(f"Langue Whisper détectée: {whisper_lang}")
|
| 123 |
+
return whisper_lang
|
| 124 |
+
|
| 125 |
+
# Si texte trop court, retourner auto
|
| 126 |
+
if len(transcribed_text) < 10:
|
| 127 |
+
print("Texte trop court, retour auto")
|
| 128 |
+
return "auto"
|
| 129 |
+
|
| 130 |
+
# Fallback avec LangDetect
|
| 131 |
+
detected_lang = detect(transcribed_text)
|
| 132 |
+
print(f"Langue LangDetect: {detected_lang}")
|
| 133 |
+
|
| 134 |
+
# Mapping des codes de langue
|
| 135 |
+
lang_mapping = {
|
| 136 |
+
'fr': 'fr', 'en': 'en', 'es': 'es', 'de': 'de', 'it': 'it',
|
| 137 |
+
'pt': 'pt', 'nl': 'nl', 'pl': 'pl', 'ru': 'ru', 'ja': 'ja',
|
| 138 |
+
'ko': 'ko', 'zh-cn': 'zh', 'zh': 'zh'
|
| 139 |
+
}
|
| 140 |
+
|
| 141 |
+
result = lang_mapping.get(detected_lang, "auto")
|
| 142 |
+
print(f"Résultat final: {result}")
|
| 143 |
+
return result
|
| 144 |
|
| 145 |
except Exception as e:
|
| 146 |
+
print(f"Erreur détection langue: {e}")
|
| 147 |
+
return "auto"
|
| 148 |
|
| 149 |
def ensure_mp3_same_name_as_input(input_path, source_wav_path):
|
| 150 |
"""
|
| 151 |
Cria um arquivo MP3 com o mesmo nome base do arquivo de entrada.
|
| 152 |
"""
|
| 153 |
try:
|
| 154 |
+
base, _ = os.path.splitext(os.path.basename(input_path))
|
| 155 |
+
mp3_path = f"{base}.mp3"
|
| 156 |
+
audio = AudioSegment.from_wav(source_wav_path)
|
| 157 |
+
audio.export(mp3_path, format="mp3", bitrate="192k")
|
| 158 |
+
return mp3_path
|
| 159 |
except Exception as e:
|
| 160 |
+
print(f"Erro ao exportar MP3: {e}")
|
| 161 |
+
return None
|
| 162 |
|
| 163 |
# --- Fonction principale ---
|
| 164 |
|
|
|
|
| 191 |
|
| 192 |
gr.Markdown("""
|
| 193 |
## ⚡ **Version optimisée**
|
| 194 |
+
- **Détection rapide** : Analyse les 7 premières secondes (hors silence initial)
|
| 195 |
- **Robuste** : Fonctionne avec tous types de fichiers
|
| 196 |
- **Timeout protection** : Évite les blocages
|
| 197 |
""")
|
| 198 |
|
| 199 |
with gr.Row():
|
| 200 |
+
with gr.Column():
|
| 201 |
+
audio_input = gr.Audio(type="filepath", label="Envoyer un fichier audio")
|
| 202 |
+
language_dropdown = gr.Dropdown(
|
| 203 |
+
choices=["auto", "fr", "en", "es", "de", "it", "pt", "nl", "pl", "ru", "ja", "ko", "zh"],
|
| 204 |
+
value="auto",
|
| 205 |
+
label="Langue (auto = détection automatique)",
|
| 206 |
+
info="Détection automatique rapide après upload"
|
| 207 |
+
)
|
| 208 |
+
submit_btn = gr.Button("Générer MP3", variant="primary")
|
| 209 |
+
reset_btn = gr.Button("Reset", variant="secondary")
|
| 210 |
+
with gr.Column():
|
| 211 |
+
language_info_output = gr.Textbox(label="Information sur la langue", lines=1)
|
| 212 |
+
mp3_download = gr.File(label="Télécharger la sortie (.mp3)")
|
| 213 |
+
mp3_playback = gr.Audio(label="Écouter la sortie (.mp3)", type="filepath")
|
| 214 |
|
| 215 |
# Détection automatique de langue lors de l'upload
|
| 216 |
audio_input.change(
|
| 217 |
+
fn=detect_language_on_upload,
|
| 218 |
+
inputs=audio_input,
|
| 219 |
+
outputs=language_dropdown
|
| 220 |
)
|
| 221 |
|
| 222 |
submit_btn.click(
|
| 223 |
+
fn=make_output_mp3,
|
| 224 |
+
inputs=[audio_input, language_dropdown],
|
| 225 |
+
outputs=[mp3_download, mp3_playback, language_info_output]
|
| 226 |
)
|
| 227 |
|
| 228 |
def reset_fields():
|
| 229 |
+
return None, None, "auto", ""
|
| 230 |
|
| 231 |
reset_btn.click(
|
| 232 |
+
fn=reset_fields,
|
| 233 |
+
inputs=[],
|
| 234 |
+
outputs=[audio_input, mp3_download, language_dropdown, language_info_output]
|
| 235 |
)
|
| 236 |
|
| 237 |
demo.launch(share=True)
|