app.py

import io
import logging
import wave
from pathlib import Path
import struct

from flask import Flask, Response, jsonify, render_template, request, send_file, stream_with_context
from flask_cors import CORS
from piper import PiperVoice

# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

app = Flask(__name__)
CORS(app)  # This will enable CORS for all routes.

# In-memory cache for PiperVoice instances
tts_instances = {}

# Directory where voice models are stored
VOICES_DIR = Path(__file__).parent / "voices"

def get_tts_instance(voice):
    """
    Retrieves a cached PiperVoice instance or creates a new one.
    Loads the model and its required .onnx.json config file.
    """
    if voice not in tts_instances:
        logger.info(f"Creating new PiperVoice instance for voice: {voice}")
        try:
            model_path, config_path = None, None
            possible_paths = [
                VOICES_DIR / f"{voice}.onnx",
                Path(__file__).parent / f"{voice}.onnx",
                Path(f"{voice}.onnx"),
            ]
            for path in possible_paths:
                if path.exists():
                    model_path = str(path)
                    potential_config_path = path.with_suffix(".onnx.json")
                    if potential_config_path.exists():
                        config_path = str(potential_config_path)
                        logger.info(f"Found model at: {model_path}")
                        logger.info(f"Found config at: {config_path}")
                        break
            
            if not model_path or not config_path:
                logger.error(f"Voice model or config not found for '{voice}'. Ensure both '.onnx' and '.onnx.json' are present.")
                return None
            
            tts_instances[voice] = PiperVoice.load(model_path, config_path=config_path)
        except Exception as e:
            logger.error(f"Failed to create PiperVoice instance for voice {voice}: {e}", exc_info=True)
            return None
    return tts_instances[voice]

@app.route('/')
def index():
    """Serves the index.html frontend."""
    return render_template('index.html')

@app.route('/api/tts', methods=['GET'])
def synthesize_audio_full():
    """
    Generates the full audio file and returns it.
    """
    text = request.args.get('text')
    voice = request.args.get('voice', 'en_GB-alba-medium')

    if not text:
        return jsonify({"error": "Text to synthesize is required."}), 400

    tts_instance = get_tts_instance(voice)
    if not tts_instance:
        return jsonify({"error": f"Could not load voice model for '{voice}'."}), 500

    try:
        wav_io = io.BytesIO()
        with wave.open(wav_io, 'wb') as wav_file:
            wav_file.setnchannels(1)
            wav_file.setsampwidth(2)  # 16-bit
            wav_file.setframerate(tts_instance.config.sample_rate)
            
            # THE CORRECT FIX, BASED ON YOUR PROVIDED `piper_tts.py`
            # The AudioChunk object has a specific attribute for the raw bytes.
            for audio_chunk in tts_instance.synthesize(text):
                wav_file.writeframes(audio_chunk.audio_int16_bytes)
        
        wav_io.seek(0)
        
        return send_file(
            wav_io,
            mimetype='audio/wav',
            as_attachment=True,
            download_name='output.wav'
        )
    except Exception as e:
        logger.error(f"Error during full synthesis: {e}", exc_info=True)
        return jsonify({"error": f"Failed to synthesize audio: {str(e)}"}), 500

def generate_audio_stream(tts_instance, text):
    """A generator function that streams the synthesized audio."""
    try:
        # 1. Create and yield the WAV header.
        def create_wav_header(sample_rate, bits_per_sample=16, channels=1):
            datasize = 2**32 - 1 # Use max value for streaming
            o = [b'RIFF', struct.pack('<I', datasize + 36), b'WAVE', b'fmt ',
                 struct.pack('<I', 16), struct.pack('<H', 1), struct.pack('<H', channels),
                 struct.pack('<I', sample_rate),
                 struct.pack('<I', sample_rate * channels * bits_per_sample // 8),
                 struct.pack('<H', channels * bits_per_sample // 8),
                 struct.pack('<H', bits_per_sample), b'data', struct.pack('<I', datasize)]
            return b"".join(o)

        header = create_wav_header(tts_instance.config.sample_rate)
        yield header
        
        # 2. THE CORRECT FIX, APPLIED TO STREAMING
        # Yield the raw bytes from the .audio_int16_bytes attribute.
        for audio_chunk in tts_instance.synthesize(text):
            yield audio_chunk.audio_int16_bytes
            
    except Exception as e:
        logger.error(f"Error during stream generation: {e}", exc_info=True)


@app.route('/api/tts-stream', methods=['GET'])
def synthesize_audio_stream():
    """
    Streams the synthesized audio back to the client as it's generated.
    """
    text = request.args.get('text')
    voice = request.args.get('voice', 'en_GB-alba-medium')

    if not text:
        return jsonify({"error": "Text to synthesize is required."}), 400

    tts_instance = get_tts_instance(voice)
    if not tts_instance:
        return jsonify({"error": f"Could not load voice model for '{voice}'."}), 500
    
    stream_generator = generate_audio_stream(tts_instance, text)
    return Response(stream_with_context(stream_generator), mimetype='audio/wav')

if __name__ == '__main__':
    app.run(debug=True, port=5001)