""" app.py - minimal web wrapper around the TTS/STT test scripts, so they can be tested from a browser instead of the command line. Built specifically for cPanel's "Setup Python App" feature (uses Passenger — see passenger_wsgi.py alongside this file, and README-CPANEL.md for the exact setup steps). Two pages: / - type Arabic text, hear it spoken (TTS test) /transcribe - upload an audio file, see it transcribed (STT test) Deliberately kept to one file, no database, no session state — this is a throwaway test tool, not something meant to become permanent. """ import os import time import wave import numpy as np from flask import Flask, request, send_file, render_template_string app = Flask(__name__) BASE_DIR = os.path.dirname(os.path.abspath(__file__)) TTS_MODEL_DIR = os.path.join(BASE_DIR, "vits-piper-ar_JO-kareem-medium") STT_MODEL_DIR = os.path.join(BASE_DIR, "sherpa-onnx-whisper-small") STT_MODEL_SIZE = "small" OUTPUT_DIR = os.path.join(BASE_DIR, "generated") os.makedirs(OUTPUT_DIR, exist_ok=True) # Models are loaded once at startup, not per-request — loading an ONNX # model from disk on every request would make each test painfully slow. _tts = None _stt = None def get_tts(): global _tts if _tts is None: import sherpa_onnx _tts = sherpa_onnx.OfflineTts( sherpa_onnx.OfflineTtsConfig( model=sherpa_onnx.OfflineTtsModelConfig( vits=sherpa_onnx.OfflineTtsVitsModelConfig( model=os.path.join(TTS_MODEL_DIR, "ar_JO-kareem-medium.onnx"), lexicon="", tokens=os.path.join(TTS_MODEL_DIR, "tokens.txt"), data_dir=os.path.join(TTS_MODEL_DIR, "espeak-ng-data"), ), num_threads=2, ), ) ) return _tts def get_stt(): global _stt if _stt is None: import sherpa_onnx _stt = sherpa_onnx.OfflineRecognizer.from_whisper( encoder=os.path.join(STT_MODEL_DIR, f"{STT_MODEL_SIZE}-encoder.int8.onnx"), decoder=os.path.join(STT_MODEL_DIR, f"{STT_MODEL_SIZE}-decoder.int8.onnx"), tokens=os.path.join(STT_MODEL_DIR, f"{STT_MODEL_SIZE}-tokens.txt"), language="ar", task="transcribe", num_threads=2, ) return _stt TTS_PAGE = """ Voice Test — Text to Speech

Voice Test — نص إلى صوت (TTS)

Switch to Speech-to-Text test →


{% if audio_file %}

Generated in {{ elapsed }}s

{% endif %} """ STT_PAGE = """ Voice Test — Speech to Text

Voice Test — صوت إلى نص (STT)

Switch to Text-to-Speech test →



{% if transcription %}
Transcribed: {{ transcription }}

Took {{ elapsed }}s

{% endif %} """ @app.route("/", methods=["GET", "POST"]) def tts_test(): audio_file = None elapsed = None text = None if request.method == "POST": text = request.form.get("text", "").strip() if text: t0 = time.time() tts = get_tts() audio = tts.generate(text, sid=0, speed=1.0) filename = f"tts_{int(time.time())}.wav" filepath = os.path.join(OUTPUT_DIR, filename) with wave.open(filepath, "wb") as f: f.setnchannels(1) f.setsampwidth(2) f.setframerate(audio.sample_rate) samples = (np.array(audio.samples) * 32767).astype(np.int16) f.writeframes(samples.tobytes()) audio_file = filename elapsed = round(time.time() - t0, 2) return render_template_string(TTS_PAGE, audio_file=audio_file, elapsed=elapsed, text=text) @app.route("/transcribe", methods=["GET", "POST"]) def stt_test(): transcription = None elapsed = None if request.method == "POST": uploaded = request.files.get("audio") if uploaded: import soundfile as sf temp_path = os.path.join(OUTPUT_DIR, f"upload_{int(time.time())}.wav") uploaded.save(temp_path) t0 = time.time() recognizer = get_stt() audio, sample_rate = sf.read(temp_path, dtype="float32") stream = recognizer.create_stream() stream.accept_waveform(sample_rate, audio) recognizer.decode_stream(stream) transcription = stream.result.text elapsed = round(time.time() - t0, 2) os.remove(temp_path) # don't accumulate uploaded audio return render_template_string(STT_PAGE, transcription=transcription, elapsed=elapsed) @app.route("/audio/") def serve_audio(filename): # Basic path-safety: only serve files from OUTPUT_DIR, only .wav if "/" in filename or ".." in filename or not filename.endswith(".wav"): return "Invalid file", 400 return send_file(os.path.join(OUTPUT_DIR, filename), mimetype="audio/wav") if __name__ == "__main__": # Only used for local testing outside cPanel — cPanel's Passenger # setup calls the `app` object directly via passenger_wsgi.py instead. app.run(debug=True, port=5000)