Update app.py
Browse files
app.py
CHANGED
|
@@ -6,11 +6,11 @@ from transformers import pipeline, VitsModel, AutoTokenizer
|
|
| 6 |
import scipy # if needed for processing
|
| 7 |
|
| 8 |
# ------------------------------------------------------
|
| 9 |
-
# 1. ASR Pipeline (English) using
|
| 10 |
# ------------------------------------------------------
|
| 11 |
asr = pipeline(
|
| 12 |
"automatic-speech-recognition",
|
| 13 |
-
model="
|
| 14 |
)
|
| 15 |
|
| 16 |
# ------------------------------------------------------
|
|
@@ -30,12 +30,13 @@ translation_tasks = {
|
|
| 30 |
|
| 31 |
# ------------------------------------------------------
|
| 32 |
# 3. TTS Model Configurations
|
| 33 |
-
#
|
| 34 |
-
#
|
|
|
|
| 35 |
# ------------------------------------------------------
|
| 36 |
tts_config = {
|
| 37 |
"Spanish": {
|
| 38 |
-
"model_id": "facebook/mms-tts-spa",
|
| 39 |
"architecture": "vits"
|
| 40 |
},
|
| 41 |
"Chinese": {
|
|
@@ -84,7 +85,7 @@ def get_tts_model(lang):
|
|
| 84 |
arch = config["architecture"]
|
| 85 |
|
| 86 |
try:
|
| 87 |
-
#
|
| 88 |
model = VitsModel.from_pretrained(model_id)
|
| 89 |
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
| 90 |
except Exception as e:
|
|
@@ -107,14 +108,15 @@ def run_tts_inference(lang, text):
|
|
| 107 |
with torch.no_grad():
|
| 108 |
output = model(**inputs)
|
| 109 |
|
| 110 |
-
# VitsModel output is typically
|
| 111 |
-
if hasattr(output, "waveform"):
|
| 112 |
-
|
| 113 |
-
else:
|
| 114 |
-
raise RuntimeError("TTS model output does not contain 'waveform'.")
|
| 115 |
|
|
|
|
| 116 |
waveform = waveform_tensor.squeeze().cpu().numpy()
|
| 117 |
-
|
|
|
|
|
|
|
| 118 |
return (sample_rate, waveform)
|
| 119 |
|
| 120 |
# ------------------------------------------------------
|
|
@@ -122,25 +124,25 @@ def run_tts_inference(lang, text):
|
|
| 122 |
# ------------------------------------------------------
|
| 123 |
def predict(audio, text, target_language):
|
| 124 |
"""
|
| 125 |
-
1. Obtain English text (
|
| 126 |
-
2. Translate English
|
| 127 |
-
3.
|
| 128 |
"""
|
| 129 |
-
# Step 1:
|
| 130 |
if text.strip():
|
| 131 |
english_text = text.strip()
|
| 132 |
elif audio is not None:
|
| 133 |
sample_rate, audio_data = audio
|
| 134 |
|
| 135 |
-
#
|
| 136 |
if audio_data.dtype not in [np.float32, np.float64]:
|
| 137 |
audio_data = audio_data.astype(np.float32)
|
| 138 |
|
| 139 |
-
#
|
| 140 |
if len(audio_data.shape) > 1 and audio_data.shape[1] > 1:
|
| 141 |
audio_data = np.mean(audio_data, axis=1)
|
| 142 |
|
| 143 |
-
# Resample to
|
| 144 |
if sample_rate != 16000:
|
| 145 |
audio_data = librosa.resample(audio_data, orig_sr=sample_rate, target_sr=16000)
|
| 146 |
|
|
@@ -150,7 +152,7 @@ def predict(audio, text, target_language):
|
|
| 150 |
else:
|
| 151 |
return "No input provided.", "", None
|
| 152 |
|
| 153 |
-
# Step 2:
|
| 154 |
translator = get_translator(target_language)
|
| 155 |
try:
|
| 156 |
translation_result = translator(english_text)
|
|
@@ -162,6 +164,7 @@ def predict(audio, text, target_language):
|
|
| 162 |
try:
|
| 163 |
sample_rate, waveform = run_tts_inference(target_language, translated_text)
|
| 164 |
except Exception as e:
|
|
|
|
| 165 |
return english_text, translated_text, f"TTS error: {e}"
|
| 166 |
|
| 167 |
return english_text, translated_text, (sample_rate, waveform)
|
|
@@ -181,12 +184,11 @@ iface = gr.Interface(
|
|
| 181 |
gr.Textbox(label="Translation (Target Language)"),
|
| 182 |
gr.Audio(label="Synthesized Speech")
|
| 183 |
],
|
| 184 |
-
title="Multimodal Language Learning Aid
|
| 185 |
description=(
|
| 186 |
-
"
|
| 187 |
-
"
|
| 188 |
-
"
|
| 189 |
-
"3. Provides synthetic speech with TTS models:\n"
|
| 190 |
),
|
| 191 |
allow_flagging="never"
|
| 192 |
)
|
|
|
|
| 6 |
import scipy # if needed for processing
|
| 7 |
|
| 8 |
# ------------------------------------------------------
|
| 9 |
+
# 1. ASR Pipeline (English) using Wav2Vec2
|
| 10 |
# ------------------------------------------------------
|
| 11 |
asr = pipeline(
|
| 12 |
"automatic-speech-recognition",
|
| 13 |
+
model="facebook/wav2vec2-base-960h"
|
| 14 |
)
|
| 15 |
|
| 16 |
# ------------------------------------------------------
|
|
|
|
| 30 |
|
| 31 |
# ------------------------------------------------------
|
| 32 |
# 3. TTS Model Configurations
|
| 33 |
+
# - Spanish: facebook/mms-tts-spa
|
| 34 |
+
# - Chinese: myshell-ai/MeloTTS-Chinese
|
| 35 |
+
# - Japanese: myshell-ai/MeloTTS-Japanese
|
| 36 |
# ------------------------------------------------------
|
| 37 |
tts_config = {
|
| 38 |
"Spanish": {
|
| 39 |
+
"model_id": "facebook/mms-tts-spa",
|
| 40 |
"architecture": "vits"
|
| 41 |
},
|
| 42 |
"Chinese": {
|
|
|
|
| 85 |
arch = config["architecture"]
|
| 86 |
|
| 87 |
try:
|
| 88 |
+
# Attempt VITS-based loading
|
| 89 |
model = VitsModel.from_pretrained(model_id)
|
| 90 |
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
| 91 |
except Exception as e:
|
|
|
|
| 108 |
with torch.no_grad():
|
| 109 |
output = model(**inputs)
|
| 110 |
|
| 111 |
+
# VitsModel output is typically `.waveform`
|
| 112 |
+
if not hasattr(output, "waveform"):
|
| 113 |
+
raise RuntimeError("TTS model output does not contain 'waveform' attribute.")
|
|
|
|
|
|
|
| 114 |
|
| 115 |
+
waveform_tensor = output.waveform
|
| 116 |
waveform = waveform_tensor.squeeze().cpu().numpy()
|
| 117 |
+
|
| 118 |
+
# Typically 16 kHz for these VITS models
|
| 119 |
+
sample_rate = 16000
|
| 120 |
return (sample_rate, waveform)
|
| 121 |
|
| 122 |
# ------------------------------------------------------
|
|
|
|
| 124 |
# ------------------------------------------------------
|
| 125 |
def predict(audio, text, target_language):
|
| 126 |
"""
|
| 127 |
+
1. Obtain English text (ASR with Wav2Vec2 or text input).
|
| 128 |
+
2. Translate English -> target_language.
|
| 129 |
+
3. TTS for that language (using configured models).
|
| 130 |
"""
|
| 131 |
+
# Step 1: English text
|
| 132 |
if text.strip():
|
| 133 |
english_text = text.strip()
|
| 134 |
elif audio is not None:
|
| 135 |
sample_rate, audio_data = audio
|
| 136 |
|
| 137 |
+
# Convert to float32 if needed
|
| 138 |
if audio_data.dtype not in [np.float32, np.float64]:
|
| 139 |
audio_data = audio_data.astype(np.float32)
|
| 140 |
|
| 141 |
+
# Stereo -> mono if needed
|
| 142 |
if len(audio_data.shape) > 1 and audio_data.shape[1] > 1:
|
| 143 |
audio_data = np.mean(audio_data, axis=1)
|
| 144 |
|
| 145 |
+
# Resample to 16k if needed
|
| 146 |
if sample_rate != 16000:
|
| 147 |
audio_data = librosa.resample(audio_data, orig_sr=sample_rate, target_sr=16000)
|
| 148 |
|
|
|
|
| 152 |
else:
|
| 153 |
return "No input provided.", "", None
|
| 154 |
|
| 155 |
+
# Step 2: Translate
|
| 156 |
translator = get_translator(target_language)
|
| 157 |
try:
|
| 158 |
translation_result = translator(english_text)
|
|
|
|
| 164 |
try:
|
| 165 |
sample_rate, waveform = run_tts_inference(target_language, translated_text)
|
| 166 |
except Exception as e:
|
| 167 |
+
# Return error info in place of audio
|
| 168 |
return english_text, translated_text, f"TTS error: {e}"
|
| 169 |
|
| 170 |
return english_text, translated_text, (sample_rate, waveform)
|
|
|
|
| 184 |
gr.Textbox(label="Translation (Target Language)"),
|
| 185 |
gr.Audio(label="Synthesized Speech")
|
| 186 |
],
|
| 187 |
+
title="Multimodal Language Learning Aid",
|
| 188 |
description=(
|
| 189 |
+
"1. Transcribes English speech using Wav2Vec2 (or takes English text).\n"
|
| 190 |
+
"2. Translates to Spanish, Chinese, or Japanese (Helsinki-NLP models).\n"
|
| 191 |
+
"3. Provides synthetic speech with TTS models.\n"
|
|
|
|
| 192 |
),
|
| 193 |
allow_flagging="never"
|
| 194 |
)
|