import gradio as gr import requests import os import time import json from dotenv import load_dotenv # --- Configuration & Constants --- load_dotenv() REPLICATE_API_TOKENS_STR = os.getenv("REPLICATE_API_TOKENS") if not REPLICATE_API_TOKENS_STR: print("WARNING: REPLICATE_API_TOKENS not found. App will not function.") REPLICATE_API_KEYS = [] else: REPLICATE_API_KEYS = [token.strip() for token in REPLICATE_API_TOKENS_STR.split(',')] MODEL_ENDPOINT = "https://api.replicate.com/v1/models/minimax/speech-02-hd/predictions" VOICE_ID_MAP = {} try: with open("voices.json", "r", encoding="utf-8") as f: VOICE_ID_MAP = json.load(f) if not VOICE_ID_MAP: print("WARNING: voices.json is empty or could not be loaded.") VOICE_ID_PRETTY_NAMES = list(VOICE_ID_MAP.keys()) DEFAULT_VOICE_PRETTY_NAME = "Friendly Person" if "Friendly Person" in VOICE_ID_PRETTY_NAMES else (VOICE_ID_PRETTY_NAMES[0] if VOICE_ID_PRETTY_NAMES else None) except FileNotFoundError: print("ERROR: voices.json not found. Please create it.") VOICE_ID_PRETTY_NAMES = [] DEFAULT_VOICE_PRETTY_NAME = None except json.JSONDecodeError: print("ERROR: voices.json is not valid JSON.") VOICE_ID_PRETTY_NAMES = [] DEFAULT_VOICE_PRETTY_NAME = None EMOTIONS = ["auto", "neutral", "happy", "sad", "angry", "fearful", "disgusted", "surprised"] SAMPLE_RATES = [8000, 16000, 22050, 24000, 32000, 44100] BITRATES = [32000, 64000, 128000, 256000] CHANNELS = ["mono", "stereo"] LANGUAGE_BOOST_OPTIONS = ["None", "English", "Chinese", "Japanese", "Korean"] current_key_index = 0 MAX_POLLING_ATTEMPTS = 60 POLL_INTERVAL = 3 def get_next_api_key(): global current_key_index if not REPLICATE_API_KEYS: return None key = REPLICATE_API_KEYS[current_key_index] current_key_index = (current_key_index + 1) % len(REPLICATE_API_KEYS) return key def generate_speech( text, pitch, speed, volume, bitrate, channel, emotion, voice_id_pretty_name, custom_voice_id, sample_rate, language_boost, english_normalization ): if not text.strip(): gr.Warning("Text input cannot be empty.") return None # Must return a value for the audio output if not REPLICATE_API_KEYS: gr.Error("No Replicate API Tokens configured. Please set REPLICATE_API_TOKENS in secrets.") return None if not VOICE_ID_MAP and not custom_voice_id.strip(): gr.Error("Voice ID configuration is missing (voices.json empty/invalid) and no custom voice ID provided.") return None actual_voice_id_to_use = "" if custom_voice_id.strip(): actual_voice_id_to_use = custom_voice_id.strip() elif voice_id_pretty_name and voice_id_pretty_name in VOICE_ID_MAP: actual_voice_id_to_use = VOICE_ID_MAP[voice_id_pretty_name] else: gr.Error(f"Selected voice '{voice_id_pretty_name}' not found in mappings and no custom ID provided.") return None payload = { "input": { "text": text, "pitch": int(pitch), "speed": float(speed), "volume": int(volume), "bitrate": int(bitrate), "channel": channel, "emotion": emotion, "voice_id": actual_voice_id_to_use, "sample_rate": int(sample_rate), "english_normalization": bool(english_normalization) } } if language_boost and language_boost.lower() != "none": payload["input"]["language_boost"] = language_boost num_keys_to_try = len(REPLICATE_API_KEYS) last_error_message_for_key = "" for i in range(num_keys_to_try): api_key = get_next_api_key() if not api_key: # Should not happen if REPLICATE_API_KEYS is populated gr.Error("Internal error: No API keys available in the cycling pool.") return None headers_post = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"} headers_get = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"} print(f"Attempting API call with key ending: ...{api_key[-4:]}. Voice ID: {actual_voice_id_to_use}") try: response = requests.post(MODEL_ENDPOINT, json=payload, headers=headers_post, timeout=30) response.raise_for_status() result = response.json() current_status = result.get("status") print(f"Initial API Response (Key ...{api_key[-4:]}): Status '{current_status}'") prediction_url = result.get("urls", {}).get("get") logs_from_initial_call = result.get("logs") polling_attempts = 0 while current_status in ["starting", "processing"] and prediction_url: if polling_attempts >= MAX_POLLING_ATTEMPTS: last_error_message_for_key = f"Polling timed out for key ...{api_key[-4:]}." print(last_error_message_for_key) result["error"] = "Polling timed out." # For local log current_status = "failed_polling_timeout" break polling_attempts += 1 time.sleep(POLL_INTERVAL) poll_response = requests.get(prediction_url, headers=headers_get, timeout=30) poll_response.raise_for_status() result = poll_response.json() current_status = result.get("status") if current_status == "succeeded": audio_url = result.get("output") if audio_url: success_logs = result.get('logs', logs_from_initial_call if logs_from_initial_call else 'N/A') print(f"Success with key ...{api_key[-4:]}. Logs: {success_logs}") gr.Info("Success! Audio generated.") return audio_url else: last_error_message_for_key = f"API succeeded (Key ...{api_key[-4:]}) but no output URL. Resp: {result}" print(last_error_message_for_key) continue # Try next key else: # Covers "failed", "failed_polling_timeout", or other unexpected states error_detail = result.get("error", f"Unknown error or unexpected status '{current_status}'") last_error_message_for_key = f"Prediction failed/timed out for key ...{api_key[-4:]}. Status: {current_status}. Error: {error_detail}" print(last_error_message_for_key) continue # Try next key except requests.exceptions.HTTPError as e: error_text = "Unknown HTTP Error" try: error_text = e.response.text except AttributeError: pass last_error_message_for_key = f"HTTP error for key ...{api_key[-4:]}: {e.response.status_code} - {error_text}" print(last_error_message_for_key) continue # Try next key except requests.exceptions.RequestException as e: last_error_message_for_key = f"Request exception for key ...{api_key[-4:]}: {e}" print(last_error_message_for_key) continue # Try next key # If all keys failed final_error_message = "All API keys failed or an unrecoverable error occurred." if last_error_message_for_key: # Provide a bit more context from the last attempt if available final_error_message += f" Last attempt error: {last_error_message_for_key}" gr.Error(final_error_message) return None # --- Gradio UI --- with gr.Blocks(theme=gr.themes.Soft()) as app: gr.Markdown("# Glue Up Academy Narrator") gr.Markdown("Enter text and adjust parameters to generate speech.") with gr.Row(): with gr.Column(scale=2): text_input = gr.Textbox( label="Text to Synthesize", lines=5, placeholder="Enter your text here...\nšŸ’”Insert '<#0.5#>' to add a 0.5s pause. Adjust duration." ) with gr.Accordion("Voice Selection", open=True): voice_id_dropdown = gr.Dropdown( label="Choose a Voice ID", choices=VOICE_ID_PRETTY_NAMES, value=DEFAULT_VOICE_PRETTY_NAME ) custom_voice_id_input = gr.Textbox( label="Custom Voice ID (Optional)", placeholder="e.g., my_cloned_voice_v2", info="If filled, this will override dropdown." ) #gr.Markdown("[Minimax Voices](https://www.minimax.io/audio/voices) for more options.") gr.Markdown("For voice cloning, reach out to Raffy") with gr.Accordion("Advanced Speech Parameters", open=False): speed_slider = gr.Slider(label="Speed", minimum=0.5, maximum=2, step=0.1, value=1.0) volume_slider = gr.Slider(label="Volume", minimum=0, maximum=10, step=1, value=1) pitch_slider = gr.Slider(label="Pitch", minimum=-12, maximum=12, step=1, value=0) english_norm_checkbox = gr.Checkbox(label="English Normalization", value=False, info="Improves number reading.") with gr.Accordion("Audio Format & Emotion", open=False): emotion_dropdown = gr.Dropdown(label="Emotion", choices=EMOTIONS, value="auto") sample_rate_dropdown = gr.Dropdown(label="Sample Rate (Hz)", choices=SAMPLE_RATES, value=32000, type="value") bitrate_dropdown = gr.Dropdown(label="Bitrate (bps)", choices=BITRATES, value=128000, type="value") channel_dropdown = gr.Dropdown(label="Channels", choices=CHANNELS, value="mono") language_boost_dropdown = gr.Dropdown(label="Language Boost", choices=LANGUAGE_BOOST_OPTIONS, value="None") with gr.Column(scale=1): generate_button = gr.Button("Generate Speech", variant="primary") # REMOVED status_message Textbox audio_output = gr.Audio(label="Generated Speech", type="filepath") generate_button.click( fn=generate_speech, inputs=[ text_input, pitch_slider, speed_slider, volume_slider, bitrate_dropdown, channel_dropdown, emotion_dropdown, voice_id_dropdown, custom_voice_id_input, sample_rate_dropdown, language_boost_dropdown, english_norm_checkbox ], outputs=[audio_output] # REMOVED status_message from outputs ) if __name__ == "__main__": if not REPLICATE_API_KEYS: print("FATAL: REPLICATE_API_TOKENS are not set.") if not VOICE_ID_MAP: print("WARNING: Voice ID map is empty (voices.json issue?).") app.launch(debug=True)