collabora / WhisperLive

WhisperLive A nearly-live implementation of OpenAI's Whisper. This project is a real-time transcription application that uses the OpenAI Whisper model to convert speech input into text output. It can be used to transcribe both live audio input from microphone and pre-recorded audio files. • Installation • Getting Started • Running the Server • Running the Client • Browser Extensions • Whisper Live Server in Docker • Future Work • Blog Posts • Contact • Citations Installation • Install PortAudio • Install whisper-live from pip • Install 3.12 venv on Fedora OpenAI REST interface Server Client Setting up NVIDIA/TensorRT-LLM for TensorRT backend • Please follow TensorRT_whisper readme for setup of NVIDIA/TensorRT-LLM and for building Whisper-TensorRT engine. Getting Started The server supports 3 backends , and . If running backend follow TensorRT_whisper readme Running the Server • Faster Whisper backend • TensorRT backend. Currently, we recommend to only use the docker setup for TensorRT. Follow TensorRT_whisper readme which works as expected. Make sure to build your TensorRT Engines before running the server with TensorRT backend. • Use option to restrict the number of clients the server should allow. Defaults to 4. • Use options to limit connection time for a client in seconds. Defaults to 600. • WhisperLive now supports the OpenVINO backend for efficient inference on Intel CPUs, iGPU and dGPUs. Currently, we tested the models uploaded to huggingface by OpenVINO. • > **Docker Recommended:** Running WhisperLive with OpenVINO inside Docker automatically enables GPU support (iGPU/dGPU) without requiring additional host setup. • > **Native (non-Docker) Use:** If you prefer running outside Docker, ensure the Intel drivers and OpenVINO runtime are installed and properly configured on your system. Refer to the documentation for installing OpenVINO. Controlling OpenMP Threads To control the number of threads used by OpenMP, you can set the environment variable. This is useful for managing CPU resources and ensuring consistent performance. If not specified, is set to by default. You can change this by using the argument: Single model mode By default, when running the server without specifying a model, the server will instantiate a new whisper model for every client connection. This has the advantage, that the server can use different model sizes, based on the client's requested model size. On the other hand, it also means you have to wait for the model to be loaded upon client connection and you will have increased (V)RAM usage. When serving a custom TensorRT model using the or a custom faster_whisper model using the option, the server will instead only instantiate the custom model once and then reuse it for all client connections. If you don't want this, set . Running the Client Use the below command to run the client: This will connect to the localhost server running on port 9090 by default. Use flags and to use different configurations. The above command will transcribe audio file provided with flag. Here are the details of client instance implemented in script: • : Language of the input audio, applicable only if using a multilingual model. • : If set to then translate from any language to . • : Whisper model size. • : Whether to use on the server. • : Set to True to save the microphone input as a file during live transcription. This option is helpful for recording sessions for later playback or analysis. Defaults to . • : Specifies the file path where the microphone input will be saved if is set to . • : Whether to mute audio playback when transcribing an audio file. Defaults to False. • : Start translation thread on the server (from any to any). • : Server translation thread's target translation language. It connects to the server running on localhost at port 9090. Using a multilingual model, language for the transcription will be automatically detected. You can also use the language option to specify the target language for the transcription, in this case, English ("en"). The translate option should be set to if we want to translate from the source language to English and if we want to transcribe in the source language. • Transcribe an audio file: • To transcribe from microphone: • To transcribe from a RTSP stream: • To transcribe from a HLS stream: Browser Extensions • Run the server with your desired backend as shown here. • Transcribe audio directly from your browser using our Chrome or Firefox extensions. Refer to Audio-Transcription-Chrome and https://github.com/collabora/WhisperLive/blob/main/TensorRT_whisper.md iOS Client Use WhisperLive on iOS with our native iOS client. Refer to and for setup and usage instructions. Whisper Live Server in Docker • GPU • Faster-Whisper • TensorRT. Refer to TensorRT_whisper readme for setup and more tensorrt backend configurations. • OpenVINO • CPU • Faster-whisper Future Work • [x] Add translation to other languages on top of transcription. Blog Posts • Transforming speech technology with WhisperLive • WhisperFusion: Ultra-low latency conversations with an AI chatbot powered by WhisperLive • Breaking language barriers 2.0: Moving closer towards fully reliable, production-ready Hindi ASR which is used in WhisperLive for hindi. Contact We are available to help you with both Open Source and proprietary AI projects. You can reach us via the Collabora website or vineet.suryan@collabora.com and marcus.edel@collabora.com. Citations `bibtex @misc{Silero VAD, author = {Silero Team}, title = {Silero VAD: pre-trained enterprise-grade Voice Activity Detector (VAD), Number Detector and Language Classifier}, year = {2021}, publisher = {GitHub}, journal = {GitHub repository}, howpublished = {\url{https://github.com/snakers4/silero-vad}}, email = {hello@silero.ai} }