server.py

from flask import Flask, request, send_file, jsonify, after_this_request
from TTS.api import TTS
import os
import uuid
import soundfile as sf
import torch

app = Flask(__name__)

MODEL_MULTI = "tts_models/multilingual/multi-dataset/xtts_v2"
tts_multi = TTS(model_name=MODEL_MULTI, gpu=False)

SPEAKERS = {
    "en_male": "Baldur Sanjin",
    "en_female": "Gracie Wise",
    "ar_male": "Damian Black", 
    "ar_female": "Claribel Dervla"
}
print("EN/AR Speakers:", SPEAKERS)

# Use high-level API only - no low-level synthesizer access
# tts.speakers lists all 58 presets

@app.route("/tts", methods=["POST"])
def tts_api():
    text = request.form.get("text", "")
    language = request.form.get("lang", "en")
    gender = request.form.get("voice", "female")
    speed = float(request.form.get("speed", 1.0))

    if not text:
        return jsonify({"error": "Text is required"}), 400
    
    speaker_name = SPEAKERS.get(f"{language}_{gender}", "Baldur Sanjin")
    
    out_path = f"/tmp/{uuid.uuid4()}.wav"

    try:
        # High-level tts_to_file with preset speaker - XTTS handles internals safely
        tts_multi.tts_to_file(
            text=text,
            speaker=speaker_name,  # Works for presets without files in recent TTS[web:26]
            language=language,
            file_path=out_path,
            speed=speed,
            split_sentences=True  # Better for speed control
        )
    except Exception as e:
        # Fallback: Provide dummy WAV for cloning mode (forces preset lookup)
        dummy_wav = "/tmp/dummy_silent.wav"
        if not os.path.exists(dummy_wav):
            sf.write(dummy_wav, torch.zeros(24000 * 3), 24000)  # 3s silence
        
        tts_multi.tts_to_file(
            text=text,
            speaker_wav=dummy_wav,  # Triggers embedding computation from preset name
            language=language,
            file_path=out_path,
            speed=speed
        )
        os.remove(dummy_wav)

    @after_this_request
    def cleanup(response):
        try:
            os.remove(out_path)
        except:
            pass
        return response

    return send_file(out_path, mimetype="audio/wav")

@app.route("/speakers", methods=["GET"])  # Bonus: List speakers
def list_speakers():
    speakers = getattr(tts_multi, 'speakers', [])
    return jsonify({"speakers": speakers[:20], "total": len(speakers)})

if __name__ == "__main__":
    app.run(host="0.0.0.0", port=7860, debug=False)