pyht is a Python SDK for Play's AI Text-to-Speech API. Play builds conversational voice AI models for realtime use cases. With pyht, you can easily convert text into high-quality audio streams with humanlike voices.
Currently the library supports only streaming and non-streaming text-to-speech. For the full set of functionalities provided by the Play API such as Voice Cloning, see the Play docs
aiohttpfilelockgrpcrequestswebsocketsYou can install the pyht SDK using pip:
pip install pyht
You can use the pyht SDK by creating a Client instance and calling its tts method. Here's a simple example:
from pyht import Client
from dotenv import load_dotenv
from pyht.client import TTSOptions
import os
load_dotenv()
client = Client(
user_id=os.getenv("PLAY_HT_USER_ID"),
api_key=os.getenv("PLAY_HT_API_KEY"),
)
options = TTSOptions(voice="s3://voice-cloning-zero-shot/775ae416-49bb-4fb6-bd45-740f205d20a1/jennifersaad/manifest.json")
for chunk in client.tts("Hi, I'm Jennifer from Play. How can I help you today?", options):
# do something with the audio chunk
print(type(chunk))
It is also possible to stream text instead of submitting it as a string all at once:
for chunk in client.stream_tts_input(some_iterable_text_stream, options):
# do something with the audio chunk
print(type(chunk))
An asyncio version of the client is also available:
from pyht import AsyncClient
client = AsyncClient(
user_id=os.getenv("PLAY_HT_USER_ID"),
api_key=os.getenv("PLAY_HT_API_KEY"),
)
options = TTSOptions(voice="s3://voice-cloning-zero-shot/775ae416-49bb-4fb6-bd45-740f205d20a1/jennifersaad/manifest.json")
async for chunk in client.tts("Hi, I'm Jennifer from Play. How can I help you today?", options):
# do something with the audio chunk
print(type(chunk))
The tts method takes the following arguments:
text: The text to be converted to speech; a string or list of strings.options: The options to use for the TTS request; a TTSOptions object (see below).voice_engine: The voice engine to use for the TTS request; a string (default Play3.0-mini-http).
PlayDialog-*: Our large, expressive multilingual model, which also supports multi-turn two-speaker dialogues.
PlayDialog-http: Streaming and non-streaming audio over HTTP.PlayDialog-ws: Streaming audio over WebSockets.Play3.0-mini-*: Our small, fast multilingual model.
Play3.0-mini-http: Streaming and non-streaming audio over HTTP.Play3.0-mini-ws: Streaming audio over WebSockets.Play3.0-mini-grpc: Streaming audio over gRPC. NOTE: This voice engine is ONLY available for Play On-Prem customers.PlayHT2.0-turbo: Our legacy English-only model, streaming audio over gRPC.streaming: Whether or not to stream the audio in chunks (default True); non-streaming is only enabled for HTTP endpoints.The TTSOptions class is used to specify the options for the TTS request. It has the following members, with these supported values:
voice: The voice to use for the TTS request; a string.
format: The format of the audio to be returned; a Format enum value.
FORMAT_MP3 (default)FORMAT_WAVFORMAT_MULAWFORMAT_FLACFORMAT_OGGFORMAT_RAWsample_rate: The sample rate of the audio to be returned; an integer or None (Play backend will choose by default).
quality: DEPRECATED (use sample rate to adjust audio quality)speed: The speed of the audio to be returned, a float (default 1.0).seed: Random seed to use for audio generation, an integer (default None, will be randomly generated).temperature (all models): The temperature of the model, a float.top_p (all models): The top_p of the model, a float.text_guidance (Play3.0-mini-* and PlayHT2.0-turbo only): The text_guidance of the model, a float.voice_guidance (Play3.0-mini-* and PlayHT2.0-turbo only): The voice_guidance of the model, a float.style_guidance (Play3.0-mini-* only): The style_guidance of the model, a float.repetition_penalty (Play3.0-mini-* and PlayHT2.0-turbo only): The repetition_penalty of the model, a float.disable_stabilization (PlayHT2.0-turbo only): Disable the audio stabilization process, a boolean (default False).language (Play3.0-* and PlayDialog-* only): The language of the text to be spoken, a Language enum value or None (default ENGLISH).
AFRIKAANSALBANIANAMHARICARABICBENGALIBULGARIANCATALANCROATIANCZECHDANISHDUTCHENGLISHFRENCHGALICIANGERMANGREEKHEBREWHINDIHUNGARIANINDONESIANITALIANJAPANESEKOREANMALAYMANDARINPOLISHPORTUGUESERUSSIANSERBIANSPANISHSWEDISHTAGALOGTHAITURKISHUKRAINIANURDUXHOSAPlayDialog-* model; if unset, the model will use default values chosen by Play.
voice_2 (multi-turn dialogue only): The second voice to use for a multi-turn TTS request; a string.
turn_prefix (multi-turn dialogue only): The prefix for the first speaker's turns in a multi-turn TTS request; a string.turn_prefix_2 (multi-turn dialogue only): The prefix for the second speaker's turns in a multi-turn TTS request; a string.voice_conditioning_seconds: How much of the voice's reference audio to pass to the model as a guide; an integer.voice_conditioning_seconds_2 (multi-turn dialogue only): How much of the second voice's reference audio to pass to the model as a guide; an integer.scene_description: A description of the overall scene (single- or multi-turn) to guide the model; a string (NOTE: currently not recommended).turn_clip_description (multi-turn dialogue only): A description of each turn (with turn prefixes) to guide the model; a string (NOTE: currently not recommended).num_candidates: How many candidates to rank to choose the best one; an integer.candidate_ranking_method: The method for the model to use to rank candidates; a CandidateRankingMethod enum value or None (let Play choose the method by default).
MeanProbRank: Rank candidates based on mean probability of the output sequence.EndProbRank: Rank candidates based on probability of end of sequence.MeanProbWithEndProbRank: Combination of MeanProbRank and EndProbRank.DescriptionRank: Rank candidates based on adherence to description.ASRRank: Rank candidates based on comparison of ASR transcription with the ground-truth text.DescriptionASRRank: Combination of DescriptionRank and ASRRank.ASRWithMeanProbRank: Combination of ASRRank and MeanProbRank.DescriptionASRWithMeanProbRank: Combination of DescriptionASRRank and MeanProbRank.You can run the provided demo from the command line.
Note: This demo depends on the following packages:
pip install numpy soundfile
python demo/main.py --user $PLAY_HT_USER_ID --key $PLAY_HT_API_KEY --text "Hello from Play!"
To run with the asyncio client, use the --async flag:
python demo/main.py --user $PLAY_HT_USER_ID --key $PLAY_HT_API_KEY --text "Hello from Play!" --async
To run with the HTTP API, which uses our latest Play3.0-mini model, use the --http flag:
python demo/main.py --user $PLAY_HT_USER_ID --key $PLAY_HT_API_KEY --text "Hello from Play!" --http
To run with the WebSockets API, which also uses our latest Play3.0-mini model, use the --ws flag:
python demo/main.py --user $PLAY_HT_USER_ID --key $PLAY_HT_API_KEY --text "Hello from Play!" --ws
The HTTP and WebSockets APIs can also be used with the async client:
python demo/main.py --user $PLAY_HT_USER_ID --key $PLAY_HT_API_KEY --text "Hello from Play!" --http --async
python demo/main.py --user $PLAY_HT_USER_ID --key $PLAY_HT_API_KEY --text "Hello from Play!" --ws --async
Alternatively, you can run the demo in interactive mode:
python demo/main.py --user $PLAY_HT_USER_ID --key $PLAY_HT_API_KEY --interactive
In interactive mode, you can input text lines to generate and play audio on-the-fly. An empty line will exit the interactive session.
To get started with the pyht SDK, you'll need your API Secret Key and User ID. Follow these steps to obtain them:
Access the API Page: Navigate to the API Access page.
Generate Your API Secret Key:
Locate Your User ID: Find and copy your User ID, which can be found on the same page under the "User ID" section.
Keep your API Secret Key confidential. It's crucial not to share it with anyone or include it in publicly accessible code repositories.
FAQs
Unknown package
We found that pyht demonstrated a healthy version release cadence and project activity because the last version was released less than a year ago. It has 1 open source maintainer collaborating on the project.
Did you know?
Socket for GitHub automatically highlights issues in each pull request and monitors the health of all your open source dependencies. Discover the contents of your packages and block harmful activity before you install or update your dependencies.
Security News
PyPI confirms no security flaws were exploited in the Ultralytics supply chain attack and highlights improvements for safer package publishing.
Security News
Research
The Socket Research Team breaks down a malicious wrapper package that uses obfuscation to harvest credentials and exfiltrate sensitive data.
Research
Security News
Attackers used a malicious npm package typosquatting a popular ESLint plugin to steal sensitive data, execute commands, and exploit developer systems.