sudhanm's picture
Update app.py
189dfd8 verified
raw
history blame
11.2 kB
import gradio as gr
import random, difflib, re, warnings, contextlib
import torch
import numpy as np
import librosa, soundfile as sf
import jiwer
# Optional transliteration
try:
from indic_transliteration import sanscript
from indic_transliteration.sanscript import transliterate
INDIC_OK = True
except:
INDIC_OK = False
# Optional HF Spaces decorator
try:
import spaces
GPU_DECORATOR = spaces.GPU
except:
class _NoOp:
def __call__(self, f): return f
GPU_DECORATOR = _NoOp()
warnings.filterwarnings("ignore")
# ---------------- CONFIG ---------------- #
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
DEVICE_INDEX = 0 if DEVICE == "cuda" else -1
DTYPE = torch.float16 if DEVICE == "cuda" else torch.float32
amp_ctx = torch.cuda.amp.autocast if DEVICE == "cuda" else contextlib.nullcontext
print(f"🔧 Using device: {DEVICE}")
LANG_CODES = {"English": "en", "Tamil": "ta", "Malayalam": "ml"}
# Primary: IndicWhisper
INDICWHISPER_MODEL = "parthiv11/indic_whisper_nodcil"
# Specialised fallbacks
SPECIALIZED_MODELS = {
"English": "openai/whisper-base.en",
"Tamil": "vasista22/whisper-tamil-large-v2",
"Malayalam": "thennal/whisper-medium-ml",
}
# Scripts and banking
SCRIPT_PATTERNS = {
"Tamil": re.compile(r"[஀-௿]"),
"Malayalam": re.compile(r"[ഀ-ൿ]"),
"English": re.compile(r"[A-Za-z]"),
}
SENTENCE_BANK = {
"English": ["The sun sets over the beautiful horizon.", "Hard work always pays off in the end."],
"Tamil": ["இன்று நல்ல வானிலை உள்ளது.", "உழைப்பு எப்போதும் வெற்றியைத் தரும்."],
"Malayalam": ["എനിക്ക് മലയാളം വളരെ ഇഷ്ടമാണ്.", "കഠിനാധ്വാനം എപ്പോഴും ഫലം നൽകും."]
}
# Model cache
indicwhisper_pipeline = None
fallback_models = {}
WHISPER_JAX_AVAILABLE = False
# ---------------- HELPERS ---------------- #
def get_random_sentence(language_choice):
return random.choice(SENTENCE_BANK[language_choice])
def is_script(text, lang_name):
p = SCRIPT_PATTERNS.get(lang_name)
return not p or bool(p.search(text or ""))
def transliterate_to_hk(text, lang_choice):
if not INDIC_OK:
return text
mapping = {"Tamil": sanscript.TAMIL, "Malayalam": sanscript.MALAYALAM, "English": None}
script = mapping.get(lang_choice)
if script and is_script(text, lang_choice):
try: return transliterate(text, script, sanscript.HK)
except: return text
return text
def preprocess_audio(audio_path, target_sr=16000):
try:
audio, sr = librosa.load(audio_path, sr=target_sr, mono=True)
if audio is None or len(audio) == 0: return None, None
audio = audio.astype(np.float32)
m = np.max(np.abs(audio))
if m > 0: audio /= m
audio, _ = librosa.effects.trim(audio, top_db=20)
if len(audio) < int(target_sr*0.1): return None, None
return audio, target_sr
except: return None, None
JIWER_TRANSFORM = jiwer.Compose([jiwer.ToLowerCase(), jiwer.RemovePunctuation(),
jiwer.RemoveMultipleSpaces(), jiwer.Strip(),
jiwer.ReduceToListOfListOfWords()])
def compute_wer(ref,hyp):
try: return jiwer.wer(ref, hyp, truth_transform=JIWER_TRANSFORM, hypothesis_transform=JIWER_TRANSFORM)
except: return 1.0
def compute_cer(ref,hyp):
try: return jiwer.cer(ref, hyp)
except: return 1.0
# ---------------- MODEL LOADERS ---------------- #
@GPU_DECORATOR
def load_indicwhisper():
global indicwhisper_pipeline, WHISPER_JAX_AVAILABLE
if indicwhisper_pipeline: return indicwhisper_pipeline
try:
from whisper_jax import FlaxWhisperPipeline; import jax.numpy as jnp
indicwhisper_pipeline = FlaxWhisperPipeline(INDICWHISPER_MODEL, dtype=jnp.bfloat16, batch_size=1)
WHISPER_JAX_AVAILABLE = True
print("✅ JAX IndicWhisper loaded!")
return indicwhisper_pipeline
except Exception as e:
print(f"⚠️ JAX unavailable: {e}"); WHISPER_JAX_AVAILABLE = False
from transformers import pipeline
indicwhisper_pipeline = pipeline("automatic-speech-recognition", model=INDICWHISPER_MODEL, device=DEVICE_INDEX)
print("✅ Transformers IndicWhisper loaded!")
return indicwhisper_pipeline
@GPU_DECORATOR
def load_specialized_model(language):
if language in fallback_models: return fallback_models[language]
from transformers import AutoProcessor, AutoModelForSpeechSeq2Seq
name = SPECIALIZED_MODELS[language]
proc = AutoProcessor.from_pretrained(name)
model = AutoModelForSpeechSeq2Seq.from_pretrained(name, torch_dtype=DTYPE).to(DEVICE)
fallback_models[language] = {"processor": proc, "model": model}
return fallback_models[language]
# ---------------- TRANSCRIBE ---------------- #
@GPU_DECORATOR
def transcribe_with_primary_model(audio_path, language):
try:
pl = load_indicwhisper(); lang_code = LANG_CODES.get(language, "en")
if WHISPER_JAX_AVAILABLE:
res = pl(audio_path, task="transcribe", language=lang_code)
if isinstance(res, dict): return res.get("text","").strip()
return str(res).strip()
if hasattr(pl, "model") and hasattr(pl, "tokenizer"):
try:
forced_ids = pl.tokenizer.get_decoder_prompt_ids(language=lang_code, task="transcribe")
pl.model.config.forced_decoder_ids = forced_ids
except: pass
with amp_ctx():
out = pl(audio_path)
if isinstance(out, dict): return (out.get("text") or "").strip()
return str(out).strip()
except Exception as e:
return f"Error: {str(e)}"
@GPU_DECORATOR
def transcribe_with_specialized_model(audio_path, language):
try:
comp = load_specialized_model(language)
audio, sr = preprocess_audio(audio_path)
if audio is None: return "Error: Audio too short"
inputs = comp["processor"](audio, sampling_rate=sr, return_tensors="pt")
feats = inputs.input_features.to(DEVICE)
gen_kwargs = {"inputs": feats, "max_length": 200, "num_beams": 3}
if language != "English":
try:
forced_ids = comp["processor"].tokenizer.get_decoder_prompt_ids(LANG_CODES[language], task="transcribe")
gen_kwargs["forced_decoder_ids"] = forced_ids
except: pass
with torch.no_grad(), amp_ctx():
ids = comp["model"].generate(**gen_kwargs)
text = comp["processor"].batch_decode(ids, skip_special_tokens=True)[0]
return text.strip()
except Exception as e:
return f"Error: {str(e)}"
@GPU_DECORATOR
def transcribe_audio(audio_path, language, use_specialized=False):
if use_specialized:
return transcribe_with_specialized_model(audio_path, language)
else:
return transcribe_with_primary_model(audio_path, language)
# ---------------- MAIN ---------------- #
@GPU_DECORATOR
def compare_pronunciation(audio, lang_choice, intended):
if audio is None: return ("❌ Please record audio first.","","","","","","","")
if not intended.strip(): return ("❌ Please generate a sentence first.","","","","","","","")
ptext = transcribe_audio(audio, lang_choice, False)
stext = transcribe_audio(audio, lang_choice, True)
actual = ptext if not ptext.startswith("Error:") else stext
if actual.startswith("Error:"): return (f"❌ {actual}","","","","","","","")
wer_val, cer_val = compute_wer(intended, actual), compute_cer(intended, actual)
score, feedback = get_score(wer_val, cer_val)
return (f"✅ Done - {score}\n💬 {feedback}",
ptext, stext,
f"{wer_val:.3f} ({(1-wer_val)*100:.1f}%)",
f"{cer_val:.3f} ({(1-cer_val)*100:.1f}%)",
diff_html(intended, actual),
char_html(intended, actual),
f"🎯 Target: {intended}")
def get_score(wer, cer):
c = (wer*0.7)+(cer*0.3)
if c <= 0.1: return "🏆 Excellent!","Outstanding!"
elif c <= 0.2: return "🎉 Very Good!","Minor improvements needed."
elif c <= 0.4: return "👍 Good!","Keep practicing."
elif c <= 0.6: return "📚 Needs Practice","Focus on clearer pronunciation."
else: return "💪 Keep Trying!","Don't give up!"
def diff_html(ref,hyp): return highlight_differences(ref,hyp)
def char_html(ref,hyp): return char_level_highlight(ref,hyp)
# Diff functions
def highlight_differences(ref,hyp):
ref_w, hyp_w = ref.split(), hyp.split()
sm = difflib.SequenceMatcher(None, ref_w, hyp_w)
out=[]
for tag,i1,i2,j1,j2 in sm.get_opcodes():
if tag=='equal': out += [f"<span style='color:green'>{w}</span>" for w in ref_w[i1:i2]]
elif tag=='replace':
out += [f"<span style='color:red'>{w}</span>" for w in ref_w[i1:i2]]
out += [f"<span style='color:orange'>→{w}</span>" for w in hyp_w[j1:j2]]
elif tag=='delete':
out += [f"<span style='color:red'>{w}</span>" for w in ref_w[i1:i2]]
elif tag=='insert':
out += [f"<span style='color:orange'>+{w}</span>" for w in hyp_w[j1:j2]]
return " ".join(out)
def char_level_highlight(ref,hyp):
sm = difflib.SequenceMatcher(None, list(ref), list(hyp))
out=[]
for tag,i1,i2,j1,j2 in sm.get_opcodes():
if tag=='equal': out += [f"<span style='color:green'>{c}</span>" for c in ref[i1:i2]]
elif tag in ('replace','delete'): out += [f"<span style='color:red'>{c}</span>" for c in ref[i1:i2]]
elif tag=='insert': out += [f"<span style='color:orange'>{c}</span>" for c in hyp[j1:j2]]
return "".join(out)
# ---------------- UI ---------------- #
def create_interface():
with gr.Blocks() as demo:
gr.Markdown("# 🎙️ IndicWhisper Pronunciation Trainer")
with gr.Row():
lang = gr.Dropdown(choices=list(LANG_CODES.keys()), value="Tamil", label="Language")
btn = gr.Button("🎲 Generate Sentence")
intended = gr.Textbox(label="Practice Sentence", interactive=False, lines=3)
audio = gr.Audio(sources=["microphone","upload"], type="filepath", label="Record")
analyze = gr.Button("🔍 Analyze")
status = gr.Textbox(label="Results", interactive=False, lines=4)
pass1 = gr.Textbox(label="Primary (IndicWhisper)")
pass2 = gr.Textbox(label="Specialized")
wer = gr.Textbox(label="Word Accuracy")
cer = gr.Textbox(label="Char Accuracy")
diff = gr.HTML(label="Word Diff")
chars = gr.HTML(label="Char Diff")
target = gr.Textbox(label="Reference", visible=False)
btn.click(get_random_sentence, [lang], [intended])
analyze.click(compare_pronunciation, [audio, lang, intended],
[status, pass1, pass2, wer, cer, diff, chars, target])
lang.change(get_random_sentence, [lang], [intended])
return demo
if __name__ == "__main__":
demo = create_interface()
demo.launch(server_name="0.0.0.0", server_port=7860, share=True)