{ "cells": [ { "cell_type": "code", "execution_count": 2, "id": "3bedf0dc-8d8e-4ede-a9e6-b8f35136aa00", "metadata": {}, "outputs": [], "source": [ "#|default_exp app" ] }, { "cell_type": "markdown", "id": "c2496690-28b2-4a79-89d5-f971b4d6f3d4", "metadata": {}, "source": [ "# Initialization" ] }, { "cell_type": "markdown", "id": "736caa6e-d79f-46e6-bf42-6594c8b809d4", "metadata": {}, "source": [ "## Get/Set Environment Variables" ] }, { "cell_type": "markdown", "id": "7baadc12-4748-4938-916f-0a256546c181", "metadata": {}, "source": [ "If you want to run this locally without having to set up the environment variables in your system, you can create a file called `tts_openai_secrets.py` in the root directory with this content:\n", "```python\n", "import os\n", "os.environ['OPENAI_API_KEY'] = 'sk-XXXXXXXXXXXXXXXXXXXXXX'\n", "os.environ['CARTESIA_API_KEY'] = 'XXXXXXXXXXXXXXXXXXXXXX'\n", "os.environ[\"ALLOWED_OAUTH_PROFILE_USERNAMES\"]= ','\n", "```" ] }, { "cell_type": "code", "execution_count": 3, "id": "667802a7-0f36-4136-a381-e66210b20462", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "OPENAI_API_KEY environment variable was not found.\n", "CARTESIA_API_KEY environment variable was not found.\n", "ALLOWED_OAUTH_PROFILE_USERNAMES environment variable was not found.\n", "import tts_openai_secrets succeeded\n" ] } ], "source": [ "#| export\n", "\n", "import os\n", "secret_import_failed = False\n", "try:\n", " # don't need the openai api key in a variable\n", " _ = os.environ['OPENAI_API_KEY']\n", " print('OPENAI_API_KEY environment variable was found.')\n", "except:\n", " print('OPENAI_API_KEY environment variable was not found.')\n", " secret_import_failed = True\n", "try:\n", " CARTESIA_API_KEY = os.environ['CARTESIA_API_KEY']\n", " print('CARTESIA_API_KEY environment variable was found.')\n", "except:\n", " print('CARTESIA_API_KEY environment variable was not found.')\n", " secret_import_failed = True\n", "try:\n", " temp_ALLOWED_OAUTH_PROFILE_USERNAMES = os.environ['ALLOWED_OAUTH_PROFILE_USERNAMES']\n", " ALLOWED_OAUTH_PROFILE_USERNAMES = tuple([o for o in temp_ALLOWED_OAUTH_PROFILE_USERNAMES.split(',') if o not in ('','None')])\n", " del temp_ALLOWED_OAUTH_PROFILE_USERNAMES\n", " print(f'ALLOWED_OAUTH_PROFILE_USERNAMES environment variable was found. {ALLOWED_OAUTH_PROFILE_USERNAMES}')\n", "except:\n", " print('ALLOWED_OAUTH_PROFILE_USERNAMES environment variable was not found.')\n", " secret_import_failed = True\n", "\n", "if secret_import_failed == True:\n", " import tts_openai_secrets\n", " _ = os.environ['OPENAI_API_KEY']\n", " CARTESIA_API_KEY = os.environ['CARTESIA_API_KEY']\n", " ALLOWED_OAUTH_PROFILE_USERNAMES = os.environ['ALLOWED_OAUTH_PROFILE_USERNAMES']\n", " print('import tts_openai_secrets succeeded')" ] }, { "cell_type": "code", "execution_count": 4, "id": "7664bc24-e8a7-440d-851d-eb16dc2d69fb", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "REQUIRE_AUTH: False\n" ] } ], "source": [ "#| export\n", "# If REQUIRE_AUTH environemnt variable is set to 'false' (from secrets) and HF_SPACE != 1 then we\n", "# are running locally and don't require authentication and authorization, otherwise we do.\n", "# We are using paid API's so don't want anybody/everybody to be able to use our paid services.\n", "if os.environ.get(\"REQUIRE_AUTH\",'true') == 'false' and os.environ.get('HF_SPACE',0) != 1:\n", " REQUIRE_AUTH = False\n", "else:\n", " REQUIRE_AUTH = True\n", "print('REQUIRE_AUTH:',REQUIRE_AUTH)" ] }, { "cell_type": "markdown", "id": "8c978095-da2a-43f8-9729-3d845e7056f1", "metadata": {}, "source": [ "## Imports" ] }, { "cell_type": "code", "execution_count": 5, "id": "4d9863fc-969e-409b-8e20-b9c3cd2cc3e7", "metadata": {}, "outputs": [], "source": [ "#| hide\n", "try:\n", " import nbdev\n", "except:\n", " print('to convert this notebook to app.py you need to pip install nbdev')" ] }, { "cell_type": "code", "execution_count": 6, "id": "4f486d3a", "metadata": {}, "outputs": [], "source": [ "#| export\n", "import os\n", "import gradio as gr\n", "import openai\n", "from pydub import AudioSegment\n", "import io\n", "from datetime import datetime\n", "from math import ceil\n", "from multiprocessing.pool import ThreadPool\n", "from functools import partial\n", "from pathlib import Path\n", "import uuid\n", "from tenacity import (\n", " retry,\n", " stop_after_attempt,\n", " wait_random_exponential,\n", ") # for exponential backoff\n", "import traceback\n", "# from cartesia.tts import CartesiaTTS\n", "import cartesia" ] }, { "cell_type": "markdown", "id": "6b425ab4-cecd-4760-84fb-b7f2cc44a565", "metadata": {}, "source": [ "Set the Gradio TEMP directory. This will be used to save audio files that were generated prior to returning them. The reason we are doing this is because if you return a bytesio object to a Gradio audio object it will not contain the file extension and will not be playable in Safari. If you pass the file to the Gradio audio object it will contain the extension. In addition if you pass the filepath instead of bytesio path, when you download the audio it will have the correct file extenion whereas otherwise it will not." ] }, { "cell_type": "markdown", "id": "852a3a1f-462a-41ab-bc94-b5ba12279ae9", "metadata": {}, "source": [ "## App Settings/Constants" ] }, { "cell_type": "code", "execution_count": 7, "id": "ecb7f207-0fc2-4d19-a313-356c05776832", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "TEMP Dir: /tmp\n" ] } ], "source": [ "#| export\n", "TEMP = os.environ.get('GRADIO_TEMP_DIR','/tmp/')\n", "TEMP_DIR = Path(TEMP)\n", "print('TEMP Dir:', TEMP_DIR)" ] }, { "cell_type": "code", "execution_count": 8, "id": "52d373be-3a79-412e-8ca2-92bb443fa52d", "metadata": {}, "outputs": [], "source": [ "#| export\n", "#Number of threads created PER USER REQUEST. This throttels the # of API requests PER USER request. This is in ADDITION to the Gradio threads.\n", "OPENAI_CLIENT_TTS_THREADS = 10 \n", "CARTESIAAI_CLIENT_TTS_THREADS = 3\n", "\n", "DEFAULT_PROVIDER = 'openai'\n", "DEFAULT_MODEL = 'tts-1'\n", "DEFAULT_VOICE = 'alloy'" ] }, { "cell_type": "code", "execution_count": 9, "id": "e5d6cac2-0dee-42d8-9b41-184b5be9cc3f", "metadata": {}, "outputs": [], "source": [ "#| export\n", "providers = dict()" ] }, { "cell_type": "code", "execution_count": 10, "id": "b77ad8d6-3289-463c-b213-1c0cc215b141", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Successfully added OpenAI as Provider\n" ] } ], "source": [ "#| export\n", "# Add OpenAI as a provider\n", "try:\n", " providers['openai'] = {\n", " 'name': 'Open AI',\n", " 'models': {o.id: o.id for o in openai.models.list().data if 'tts' in o.id},\n", " 'voices': {o:{'id':o,'name':o.title()} for o in ['alloy', 'echo', 'fable', 'onyx', 'nova', 'shimmer']},\n", " 'settings': {'max_chunk_size': 4000, 'chunk_processing_time': 60, \n", " 'threads': OPENAI_CLIENT_TTS_THREADS,\n", " 'audio_file_conversion_kwargs':{'format': 'mp3'}},\n", " }\n", " print('Successfully added OpenAI as Provider')\n", "except Exception as e:\n", " print(f\"\"\"Error: Failed to add OpenAI as a provider.\\nException: {repr(e)}\\nTRACEBACK:\\n\"\"\",traceback.format_exc())\n", "# providers" ] }, { "cell_type": "code", "execution_count": 11, "id": "87fca48b-a16a-4d2b-919c-75e88e4e5eb5", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Successfully added Cartesia AI as Provider\n" ] } ], "source": [ "#| export\n", "# Add Cartesia AI as a provider\n", "try:\n", " providers['cartesiaai'] = {\n", " 'name': 'Cartesia AI',\n", " 'models': {'upbeat-moon': 'Sonic Turbo English'},\n", " 'voices': {v['id']:v for k,v in cartesia.tts.CartesiaTTS().get_voices().items()},\n", " 'settings': {'max_chunk_size': 500, 'chunk_processing_time': 20, \n", " 'threads': CARTESIAAI_CLIENT_TTS_THREADS,\n", " 'audio_file_conversion_kwargs':{'format': 'raw', 'frame_rate': 44100, \n", " 'channels': 1, 'sample_width': 2}},\n", " }\n", " print('Successfully added Cartesia AI as Provider')\n", "except Exception as e:\n", " print(f\"\"\"Error: Failed to add Cartesia AI as a provider.\\nException: {repr(e)}\\nTRACEBACK:\\n\"\"\",traceback.format_exc())\n", "# providers" ] }, { "cell_type": "markdown", "id": "6bd2e9ed-9dbd-4d5f-a814-2942108b5935", "metadata": {}, "source": [ "EXAMPLE: providers\n", "```python\n", "{'openai': {'name': 'Open AI',\n", " 'models': {'tts-1-hd-1106': 'tts-1-hd-1106',\n", " 'tts-1-hd': 'tts-1-hd',\n", " 'tts-1': 'tts-1',\n", " 'tts-1-1106': 'tts-1-1106'},\n", " 'voices': {'alloy': {'id': 'alloy', 'name': 'Alloy'},\n", " 'echo': {'id': 'echo', 'name': 'Echo'},\n", " 'fable': {'id': 'fable', 'name': 'Fable'},\n", " 'onyx': {'id': 'onyx', 'name': 'Onyx'},\n", " 'nova': {'id': 'nova', 'name': 'Nova'},\n", " 'shimmer': {'id': 'shimmer', 'name': 'Shimmer'}}},\n", " 'cartesiaai': {'name': 'Cartesia AI',\n", " 'models': {'upbeat-moon': 'Sonic Turbo English'},\n", " 'voices': {'3b554273-4299-48b9-9aaf-eefd438e3941': {'id': '3b554273-4299-48b9-9aaf-eefd438e3941',\n", " 'user_id': None,\n", " 'is_public': True,\n", " 'name': 'Indian Lady',\n", " 'description': 'This voice is young, rich, and curious, perfect for a narrator or fictional character',\n", " 'created_at': '2024-05-04T18:48:17.006441-07:00',\n", " 'embedding': [0.015546328,-0.11384969,0.14146514, ...]},\n", " '63ff761f-c1e8-414b-b969-d1833d1c870c': {'id': '63ff761f-c1e8-414b-b969-d1833d1c870c',\n", " 'user_id': None,\n", " 'is_public': True,\n", " 'name': 'Confident British Man',\n", " 'description': 'This voice is disciplined with a British accent, perfect for a commanding character or narrator',\n", " 'created_at': '2024-05-04T18:57:31.399193-07:00',\n", " 'embedding': [-0.056990184,-0.06531749,-0.05618861,...]}\n", " }\n", "}\n", "```" ] }, { "cell_type": "code", "execution_count": 12, "id": "d1352f28-f761-4e91-a9bc-4efe47552f4d", "metadata": {}, "outputs": [], "source": [ "# {v['id']:v['name'] for k,v in cartesia.tts.CartesiaTTS().get_voices().items()}," ] }, { "cell_type": "markdown", "id": "a06e16b3-6310-462e-8192-b65ca324d86f", "metadata": {}, "source": [ "```\n", "({'3b554273-4299-48b9-9aaf-eefd438e3941': 'Indian Lady',\n", " '63ff761f-c1e8-414b-b969-d1833d1c870c': 'Confident British Man',\n", " 'daf747c6-6bc2-4083-bd59-aa94dce23f5d': 'Middle Eastern Woman',\n", " 'ed81fd13-2016-4a49-8fe3-c0d2761695fc': 'Sportsman',\n", " 'f114a467-c40a-4db8-964d-aaba89cd08fa': 'Yogaman',\n", " 'c45bc5ec-dc68-4feb-8829-6e6b2748095d': 'Movieman',\n", " '87748186-23bb-4158-a1eb-332911b0b708': 'Wizardman',\n", " '98a34ef2-2140-4c28-9c71-663dc4dd7022': 'Southern Man',\n", " '79f8b5fb-2cc8-479a-80df-29f7a7cf1a3e': 'Nonfiction Man',\n", " '36b42fcb-60c5-4bec-b077-cb1a00a92ec6': 'Pilot over Intercom',\n", " '69267136-1bdc-412f-ad78-0caad210fb40': 'Friendly Reading Man',\n", " '15a9cd88-84b0-4a8b-95f2-5d583b54c72e': 'Reading Lady',\n", " '95856005-0332-41b0-935f-352e296aa0df': 'Classy British Man',\n", " 'd46abd1d-2d02-43e8-819f-51fb652c1c61': 'Newsman',\n", " '2ee87190-8f84-4925-97da-e52547f9462c': 'Child',\n", " 'c2ac25f9-ecc4-4f56-9095-651354df60c0': 'Commercial Lady',\n", " '5345cf08-6f37-424d-a5d9-8ae1101b9377': 'Maria',\n", " 'a3520a8f-226a-428d-9fcd-b0a4711a6829': 'Reflective Woman',\n", " 'e3827ec5-697a-4b7c-9704-1a23041bbc51': 'Sweet Lady',\n", " 'a0e99841-438c-4a64-b679-ae501e7d6091': 'Barbershop Man',\n", " 'cd17ff2d-5ea4-4695-be8f-42193949b946': 'Meditation Lady',\n", " 'bf991597-6c13-47e4-8411-91ec2de5c466': 'Newslady',\n", " '41534e16-2966-4c6b-9670-111411def906': \"1920's Radioman\",\n", " '79a125e8-cd45-4c13-8a67-188112f4dd22': 'British Lady',\n", " 'a167e0f3-df7e-4d52-a9c3-f949145efdab': 'Ted',\n", " '248be419-c632-4f23-adf1-5324ed7dbf1d': 'Hannah',\n", " 'c8605446-247c-4d39-acd4-8f4c28aa363c': 'Wise Lady',\n", " '00a77add-48d5-4ef6-8157-71e5437b282d': 'Calm Lady',\n", " '638efaaa-4d0c-442e-b701-3fae16aad012': 'Indian Man',\n", " 'd6b6d712-6030-4420-9771-8b35f326fd20': 'Mat 1'},)\n", "```" ] }, { "cell_type": "code", "execution_count": 13, "id": "8eb7e7d5-7121-4762-b8d1-e5a9539e2b36", "metadata": {}, "outputs": [], "source": [ "#| export\n", "clean_text_prompt = \"\"\"Your job is to clean up text that is going to be fed into a text to speech (TTS) model. You must remove parts of the text that would not normally be spoken such as reference marks `[1]`, spurious citations such as `(Reddy et al., 2021; Wu et al., 2022; Chang et al., 2022; Kondratyuk et al., 2023)` and any other part of the text that is not normally spoken. Please also clean up sections and headers so they are on new lines with proper numbering. You must also clean up any math formulas that are salvageable from being copied from a scientific paper. If they are garbled and do not make sense then remove them. You must carefully perform the text cleanup so it is translated into speech that is easy to listen to however you must not modify the text otherwise. It is critical that you repeat all of the text without modifications except for the cleanup activities you've been instructed to do. Also you must clean all of the text you are given, you may not omit any of it or stop the cleanup task early.\"\"\"\n" ] }, { "cell_type": "markdown", "id": "e6400d8e-49e8-41b8-ad0e-18bc032682b6", "metadata": {}, "source": [ "# Main Implementation" ] }, { "cell_type": "code", "execution_count": 14, "id": "b5b29507-92bc-453d-bcc5-6402c17e9a0d", "metadata": {}, "outputs": [], "source": [ "#| export\n", "def verify_authorization(profile: gr.OAuthProfile=None) -> str:\n", " print('Profile:', profile)\n", " if REQUIRE_AUTH == False:\n", " return 'WARNING_NO_AUTH_REQUIRED_LOCAL'\n", " elif profile is not None and profile.username in ALLOWED_OAUTH_PROFILE_USERNAMES:\n", " return f\"{profile.username}\"\n", " else:\n", " # print('Unauthorized',profile)\n", " raise PermissionError(f'Your huggingface username ({profile}) is not authorized. Must be set in ALLOWED_OAUTH_PROFILE_USERNAMES environment variable.')\n", " return None" ] }, { "cell_type": "code", "execution_count": 15, "id": "24674094-4d47-4e48-b591-55faabcff8df", "metadata": {}, "outputs": [], "source": [ "#| export\n", "def split_text(input_text, provider):\n", " settings = providers[provider]['settings']\n", " max_length = settings['max_chunk_size']\n", " lookback = max_length // 4\n", " \n", " # If the text is shorter than the max_length, return it as is\n", " if len(input_text) <= max_length:\n", " return [input_text]\n", "\n", " chunks = []\n", " while input_text:\n", " # Check if the remaining text is shorter than the max_length\n", " if len(input_text) <= max_length:\n", " chunks.append(input_text)\n", " break\n", "\n", " # Define the split point, initially set to max_length\n", " split_point = max_length\n", "\n", " # Look for a newline in the last 'lookback' characters\n", " newline_index = input_text.rfind('\\n', max_length-lookback, max_length)\n", " if newline_index != -1:\n", " split_point = newline_index + 1 # Include the newline in the current chunk\n", "\n", " # If no newline, look for a period followed by space\n", " elif '. ' in input_text[max_length-lookback:max_length]:\n", " # Find the last '. ' in the lookback range\n", " period_index = input_text.rfind('. ', max_length-lookback, max_length)\n", " split_point = period_index + 2 # Split after the space\n", "\n", " # Split the text and update the input_text\n", " chunks.append(input_text[:split_point])\n", " input_text = input_text[split_point:]\n", "\n", " return chunks" ] }, { "cell_type": "code", "execution_count": 16, "id": "e6224ae5-3792-42b2-8392-3abd42998a50", "metadata": {}, "outputs": [], "source": [ "#| export\n", "def concatenate_audio(files:list, **kwargs):\n", "\n", " # Initialize an empty AudioSegment object for concatenation\n", " combined = AudioSegment.empty()\n", "\n", " # Loop through the list of mp3 binary data\n", " for data in files:\n", " # Convert binary data to an audio segment\n", " audio_segment = AudioSegment.from_file(io.BytesIO(data), **kwargs)\n", " \n", " # Concatenate this segment to the combined segment\n", " combined += audio_segment\n", "\n", " #### Return Bytes Method\n", " # # Export the combined segment to a new mp3 file\n", " # # Use a BytesIO object to handle this in memory\n", " # combined_mp3 = io.BytesIO()\n", " # combined.export(combined_mp3, format=\"mp3\")\n", "\n", " # # Seek to the start so it's ready for reading\n", " # combined_mp3.seek(0)\n", "\n", " # return combined_mp3.getvalue()\n", "\n", " #### Return Filepath Method\n", " filepath = TEMP_DIR/(str(uuid.uuid4())+'.mp3')\n", " combined.export(filepath, format=\"mp3\")\n", " print('Saving mp3 file to temp directory: ', filepath)\n", " return str(filepath)" ] }, { "cell_type": "code", "execution_count": 17, "id": "4691703d-ed0f-4481-8006-b2906289b780", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Added create_speech_func for openai provider\n" ] } ], "source": [ "#| export\n", "def create_speech_openai(chunk_idx, input, model='tts-1', voice='alloy', speed=1.0, **kwargs):\n", " client = openai.OpenAI()\n", " \n", " @retry(wait=wait_random_exponential(min=1, max=180), stop=stop_after_attempt(6))\n", " def _create_speech_with_backoff(**kwargs):\n", " return client.audio.speech.create(**kwargs)\n", " \n", " response = _create_speech_with_backoff(input=input, model=model, voice=voice, speed=speed, **kwargs)\n", " client.close()\n", " return chunk_idx, response.content\n", "if 'openai' in providers.keys():\n", " providers['openai']['settings']['create_speech_func'] = create_speech_openai\n", " print('Added create_speech_func for openai provider')" ] }, { "cell_type": "markdown", "id": "6b1a0a8a-0ff6-44fa-b85c-c80c56bd3a24", "metadata": {}, "source": [ "```python\n", "client.generate(\n", " *,\n", " transcript: str,\n", " voice: List[float],\n", " model_id: str = '',\n", " duration: int = None,\n", " chunk_time: float = None,\n", " stream: bool = False,\n", " websocket: bool = True,\n", " output_format: Union[str, cartesia._types.AudioOutputFormat] = 'fp32',\n", " data_rtype: str = 'bytes',\n", ") -> Union[cartesia._types.AudioOutput, Generator[cartesia._types.AudioOutput, NoneType, NoneType]]\n", "\n", "list(cartesia._types.AudioOutputFormat)\n", "[,\n", " ,\n", " ,\n", " ,\n", " ,\n", " ,\n", " ,\n", " ,\n", " ]\n", "```" ] }, { "cell_type": "code", "execution_count": 18, "id": "3420c868-71cb-4ac6-ac65-6f02bfd841d1", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Added create_speech_func for create_speech_cartesiaai provider\n" ] } ], "source": [ "#| export\n", "def create_speech_cartesiaai(chunk_idx, input, model='upbeat-moon', \n", " voice='248be419-c632-4f23-adf1-5324ed7dbf1d', #Hannah\n", " websocket=False, \n", " output_format='pcm_44100', \n", " **kwargs):\n", " client = cartesia.tts.CartesiaTTS()\n", " \n", " # @retry(wait=wait_random_exponential(min=1, max=180), stop=stop_after_attempt(6))\n", " def _create_speech_with_backoff(**kwargs):\n", " return client.generate(**kwargs)\n", " \n", " response = _create_speech_with_backoff(transcript=input, model_id=model, \n", " voice=client.get_voice_embedding(voice_id=voice), \n", " websocket=websocket, \n", " output_format=output_format, \n", " **kwargs)\n", " client.close()\n", " return chunk_idx, response[\"audio\"]\n", "if 'cartesiaai' in providers.keys():\n", " providers['cartesiaai']['settings']['create_speech_func'] = create_speech_cartesiaai\n", " print('Added create_speech_func for create_speech_cartesiaai provider')" ] }, { "cell_type": "code", "execution_count": 19, "id": "d0082383-9d03-4b25-b68a-080d0b28caa9", "metadata": {}, "outputs": [], "source": [ "# test\n", "# create_speech_cartesiaai(1,\"Hi. What's your name?\", model='upbeat-moon',\n", "# voice='63ff761f-c1e8-414b-b969-d1833d1c870c')\n", "# text = \"\"\"Ladies and Gentlemen, gather 'round your radio sets for a special report on a remarkable phenomenon sweeping the nation! It seems a new generation of prodigies is lighting up our schools and captivating communities from coast to coast. Yes, we're talking about gifted children, those bright young minds whose talents and intellects are far beyond their tender years.\n", "# In classrooms across the country, educators are reporting an unprecedented wave of exceptional youngsters. These gifted children, some as young as five or six, are astonishing their teachers with feats of arithmetic, literacy, and creativity that rival those of much older students. The age of enlightenment is upon us, and it starts with these brilliant boys and girls!\n", "# Take young Tommy Caldwell of Newark, New Jersey, for instance. At the ripe age of seven, Tommy has already mastered algebra and is now diving into the wonders of geometry. His teacher, Miss Grace Whittaker, says she’s never seen such a gifted mathematician in her twenty years of teaching. And Tommy is just one of many!\n", "# Then there's little Sally Henderson of Topeka, Kansas. At the tender age of eight, Sally has penned her first book of poetry, a collection that has local literary circles abuzz with excitement. Her vivid imagination and eloquent verse have earned her the nickname “The Young Bard of Topeka.” The town is swelling with pride for their young poetess.\n", "# Experts from esteemed institutions like Harvard and Yale are paying close attention to this trend, conducting studies to understand the roots of such exceptional abilities. Some suggest it’s the result of improved nutrition and health care, while others believe it’s the progressive educational methods now being employed in our schools. Regardless of the cause, one thing is clear: America is home to a new generation of geniuses!\n", "# Communities are rallying to support these young prodigies, with special programs and schools being established to nurture their extraordinary talents. Parents, teachers, and civic leaders are united in their mission to provide the resources and opportunities these gifted children need to reach their full potential.\n", "# So, dear listeners, as you sit by your radios tonight, take heart in the knowledge that our future is bright. With these remarkable young minds leading the way, the possibilities are endless. This is your announcer signing off, wishing you all a splendid evening and a hopeful tomorrow.\"\"\"\n", "# _, test_data = create_speech_cartesiaai(1, text, model='upbeat-moon',\n", "# voice='41534e16-2966-4c6b-9670-111411def906')" ] }, { "cell_type": "code", "execution_count": 20, "id": "649d90a5-9398-4cb5-a1e8-a464d463a11c", "metadata": {}, "outputs": [], "source": [ "# test_file = AudioSegment.from_file(io.BytesIO(test_data), **{'format': 'raw', 'frame_rate': 44100, 'channels': 1, 'sample_width': 2})\n", "# test_file.export('./test3.mp3', format=\"mp3\")" ] }, { "cell_type": "code", "execution_count": 21, "id": "e34bb4aa-698c-4452-8cda-bd02b38f7122", "metadata": {}, "outputs": [], "source": [ "#| export\n", "def create_speech(input_text, provider, model='tts-1', voice='alloy', \n", " profile: gr.OAuthProfile|None=None, # comment out of running locally\n", " progress=gr.Progress(), **kwargs):\n", "\n", " #Verify auth if it is required. This is very important if this is in a HF space. DO NOT DELETE!!!\n", " if REQUIRE_AUTH: verify_authorization(profile)\n", " start = datetime.now()\n", " \n", " settings = providers[provider]['settings']\n", " create_speech_func = settings['create_speech_func']\n", " chunk_processing_time = settings['chunk_processing_time']\n", " threads = settings['threads']\n", " audio_file_conversion_kwargs = settings['audio_file_conversion_kwargs']\n", " \n", " # Split the input text into chunks\n", " chunks = split_text(input_text, provider=provider)\n", "\n", " # Initialize the progress bar\n", " progress(0, desc=f\"Started processing {len(chunks)} text chunks using {threads} threads. ETA is ~{ceil(len(chunks)/threads)*chunk_processing_time/60.} min.\")\n", "\n", " # Initialize a list to hold the audio data of each chunk\n", " audio_data = []\n", "\n", " # Process each chunk\n", " with ThreadPool(processes=threads) as pool:\n", " results = pool.starmap(\n", " partial(create_speech_func, model=model, voice=voice, **kwargs), \n", " zip(range(len(chunks)),chunks)\n", " )\n", " audio_data = [o[1] for o in sorted(results)]\n", "\n", " # Progress\n", " progress(.9, desc=f\"Merging audio chunks... {(datetime.now()-start).seconds} seconds to process.\")\n", " \n", " # Concatenate the audio data from all chunks\n", " combined_audio = concatenate_audio(audio_data, **audio_file_conversion_kwargs)\n", "\n", " # Final update to the progress bar\n", " progress(1, desc=f\"Processing completed... {(datetime.now()-start).seconds} seconds to process.\")\n", " \n", " print(f\"Processing time: {(datetime.now()-start).seconds} seconds.\")\n", "\n", " return combined_audio\n" ] }, { "cell_type": "code", "execution_count": 22, "id": "ca2c6f8c-62ed-4ac1-9c2f-e3b2bfb47e8d", "metadata": {}, "outputs": [], "source": [ "# create_speech(\"Hi. What's your name?\", provider='openai', model='tts-1', voice='alloy')\n", "# create_speech(\"Hi. What's your name?\", provider='cartesiaai', model='upbeat-moon',\n", "# voice='63ff761f-c1e8-414b-b969-d1833d1c870c')" ] }, { "cell_type": "code", "execution_count": 23, "id": "236dd8d3-4364-4731-af93-7dcdec6f18a1", "metadata": {}, "outputs": [], "source": [ "#| export\n", "def get_input_text_len(input_text):\n", " return len(input_text)" ] }, { "cell_type": "code", "execution_count": 24, "id": "0523a158-ee07-48b3-9350-ee39d4deee7f", "metadata": {}, "outputs": [], "source": [ "#| export\n", "def get_generation_cost(input_text, tts_model_dropdown, provider):\n", " text_len = len(input_text)\n", " if provider == 'openai':\n", " if tts_model_dropdown.endswith('-hd'):\n", " cost = text_len/1000 * 0.03\n", " else:\n", " cost = text_len/1000 * 0.015\n", " elif provider == 'cartesiaai':\n", " cost = text_len/1000 * 0.065\n", " else:\n", " raise ValueError(f'Invalid argument provider: {provider}')\n", " return \"${:,.3f}\".format(cost)" ] }, { "cell_type": "code", "execution_count": 25, "id": "f4d1ba0b-6960-4e22-8dba-7de70370753a", "metadata": {}, "outputs": [], "source": [ "#| export\n", "def get_model_choices(provider):\n", " return sorted([(v,k) for k,v in providers[provider]['models'].items()])" ] }, { "cell_type": "code", "execution_count": 26, "id": "efa28cf2-548d-439f-bf2a-21a5edbf9eba", "metadata": {}, "outputs": [], "source": [ "#| export\n", "def update_model_choices(provider):\n", " choices = get_model_choices(provider)\n", " return gr.update(choices=choices,value=choices[0][1])" ] }, { "cell_type": "code", "execution_count": 27, "id": "cdc1dde5-5edd-4dbf-bd11-30eb418c571d", "metadata": {}, "outputs": [], "source": [ "#| export\n", "def get_voice_choices(provider, model):\n", " return sorted([(v['name'],v['id']) for v in providers[provider]['voices'].values()])" ] }, { "cell_type": "code", "execution_count": 28, "id": "035c33dd-c8e6-42b4-91d4-6bc5f1b36df3", "metadata": {}, "outputs": [], "source": [ "#| export\n", "def update_voice_choices(provider, model):\n", " choices = get_voice_choices(provider, model)\n", " return gr.update(choices=choices,value=choices[0][1])" ] }, { "cell_type": "code", "execution_count": 29, "id": "c97c03af-a377-42e1-93e0-1df957c0e4cc", "metadata": {}, "outputs": [], "source": [ "#| export\n", "def split_text_as_md(*args, **kwargs):\n", " output = split_text(*args, **kwargs)\n", " return '# Text Splits:\\n' + '
----------
'.join(output)" ] }, { "cell_type": "code", "execution_count": 30, "id": "e4fb3159-579b-4271-bc96-4cd1e2816eca", "metadata": {}, "outputs": [], "source": [ "#| export\n", "with gr.Blocks(title='TTS', head='TTS', delete_cache=(3600,3600)) as app:\n", " \n", " ### Define UI ###\n", " gr.Markdown(\"# TTS\")\n", " gr.Markdown(\"\"\"Start typing below and then click **Go** to create the speech from your text.\n", "For requests longer than allowed by the API they will be broken into chunks automatically. [Spaces Link](https://matdmiller-tts-openai.hf.space/) | Spaces Link HTML\"\"\")\n", " with gr.Row():\n", " input_text = gr.Textbox(max_lines=100, label=\"Enter text here\")\n", " with gr.Row():\n", " tts_provider_dropdown = gr.Dropdown(value=DEFAULT_PROVIDER,\n", " choices=tuple([(v['name'],k) for k,v in providers.items()]), label='Provider', interactive=True)\n", " tts_model_dropdown = gr.Dropdown(value=DEFAULT_MODEL,choices=get_model_choices(DEFAULT_PROVIDER), \n", " label='Model', interactive=True)\n", " tts_voice_dropdown = gr.Dropdown(value=DEFAULT_VOICE,choices=get_voice_choices(DEFAULT_PROVIDER, DEFAULT_MODEL),\n", " label='Voice', interactive=True)\n", " input_text_length = gr.Label(label=\"Number of characters\")\n", " generation_cost = gr.Label(label=\"Generation cost\")\n", " with gr.Row():\n", " output_audio = gr.Audio()\n", " go_btn = gr.Button(\"Go\")\n", " clear_btn = gr.Button('Clear')\n", " if REQUIRE_AUTH:\n", " gr.LoginButton()\n", " auth_md = gr.Markdown('')\n", " \n", " chunks_md = gr.Markdown('',label='Chunks')\n", " \n", "\n", " ### Define UI Actions ###\n", " \n", " # input_text \n", " input_text.input(fn=get_input_text_len, inputs=input_text, outputs=input_text_length)\n", " input_text.input(fn=get_generation_cost, \n", " inputs=[input_text,tts_model_dropdown,tts_provider_dropdown], \n", " outputs=generation_cost)\n", " input_text.input(fn=split_text_as_md, inputs=[input_text,tts_provider_dropdown], outputs=chunks_md)\n", " \n", " # tts_provider_dropdown\n", " tts_provider_dropdown.change(fn=update_model_choices, inputs=[tts_provider_dropdown], \n", " outputs=tts_model_dropdown)\n", " tts_provider_dropdown.change(fn=update_voice_choices, inputs=[tts_provider_dropdown, tts_model_dropdown], \n", " outputs=tts_voice_dropdown)\n", " tts_provider_dropdown.change(fn=split_text_as_md, inputs=[input_text,tts_provider_dropdown], outputs=chunks_md)\n", " \n", " # tts_model_dropdown\n", " tts_model_dropdown.change(fn=get_generation_cost, \n", " inputs=[input_text,tts_model_dropdown,tts_provider_dropdown], outputs=generation_cost)\n", " \n", " \n", " go_btn.click(fn=create_speech, \n", " inputs=[input_text, tts_provider_dropdown, tts_model_dropdown, tts_voice_dropdown], \n", " outputs=[output_audio])\n", " \n", " \n", " clear_btn.click(fn=lambda: '', outputs=input_text)\n", "\n", " if REQUIRE_AUTH:\n", " app.load(verify_authorization, None, auth_md)\n", " \n", " " ] }, { "cell_type": "code", "execution_count": 31, "id": "a00648a1-891b-470b-9959-f5d502055713", "metadata": {}, "outputs": [], "source": [ "#| export\n", "# launch_kwargs = {'auth':('username',GRADIO_PASSWORD),\n", "# 'auth_message':'Please log in to Mat\\'s TTS App with username: username and password.'}\n", "launch_kwargs = {}\n", "queue_kwargs = {'default_concurrency_limit':10}" ] }, { "cell_type": "code", "execution_count": 32, "id": "4b534fe7-4337-423e-846a-1bdb7cccc4ea", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Running on local URL: http://127.0.0.1:7860\n", "\n", "To create a public link, set `share=True` in `launch()`.\n" ] }, { "data": { "text/html": [ "
" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "text/plain": [] }, "execution_count": 32, "metadata": {}, "output_type": "execute_result" } ], "source": [ "#| hide\n", "#Notebook launch\n", "app.queue(**queue_kwargs)\n", "app.launch(**launch_kwargs)" ] }, { "cell_type": "code", "execution_count": null, "id": "cb886d45", "metadata": {}, "outputs": [], "source": [ "#| export\n", "#.py launch\n", "if __name__ == \"__main__\":\n", " app.queue(**queue_kwargs)\n", " app.launch(**launch_kwargs)" ] }, { "cell_type": "code", "execution_count": 33, "id": "28e8d888-e790-46fa-bbac-4511b9ab796c", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Closing server running on port: 7860\n" ] } ], "source": [ "#| hide\n", "app.close()" ] }, { "cell_type": "code", "execution_count": null, "id": "afbc9699-4d16-4060-88f4-cd1251754cbd", "metadata": {}, "outputs": [], "source": [ "#| hide\n", "gr.close_all()" ] }, { "cell_type": "code", "execution_count": 34, "id": "0420310d-930b-4904-8bd4-3458ad8bdbd3", "metadata": {}, "outputs": [], "source": [ "#| hide\n", "import nbdev\n", "nbdev.export.nb_export('app.ipynb',lib_path='.')" ] }, { "cell_type": "code", "execution_count": null, "id": "9869749d-bc7c-4e24-9dbc-403f665d6200", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "4b4fe405-b58a-471f-9ce9-e52104012409", "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "kernelspec": { "display_name": "gradio1", "language": "python", "name": "gradio1" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.11.8" } }, "nbformat": 4, "nbformat_minor": 5 }