Search Results for "inference"

AudioCraft is a PyTorch library for text-to-audio and text-to-music generation, packaging research models and tooling for training and inference. It includes MusicGen for music generation conditioned on text (and optionally melody) and AudioGen for text-conditioned sound effects and environmental audio. Both models operate over discrete audio tokens produced by a neural codec (EnCodec), which acts like a tokenizer for waveforms and enables efficient sequence modeling. The repo provides inference scripts, checkpoints, and simple Python APIs so you can generate clips from prompts or incorporate the models into applications. ...

LiveAvatar is an open-source research and implementation project that provides a unified framework for real-time, streaming, interactive avatar video generation driven by audio and other control signals. It implements techniques from state-of-the-art diffusion-based avatar modeling to support infinite-length continuous video generation with low latency, enabling interactive AI avatars that maintain continuity and realism over extended sessions. The project co-designs algorithms and system...

NovaSR is an extremely lightweight and high-performance audio upsampling model that transforms low-quality 16 kHz audio into clearer, high-fidelity 48 kHz audio with remarkable speed and efficiency. At only about 50 KB in size, the model is orders of magnitude smaller than typical audio super-resolution networks, yet it achieves high quality and realtime performance thanks to its compact architecture and efficient convolutional design. NovaSR is especially valuable for post-processing tasks...

...Mimi processes 24 kHz audio, down to a 12.5 Hz representation with a bandwidth of 1.1 kbps, in a fully streaming manner (latency of 80ms, the frame size), yet performs better than existing, non-streaming, codecs like SpeechTokenizer (50 Hz, 4kbps), or SemantiCodec (50 Hz, 1.3kbps). Moshi models two streams of audio: one corresponds to Moshi, and the other one to the user. At inference, the stream from the user is taken from the audio input, and the one for Moshi is sampled from the model's output. Along these two audio streams, Moshi predicts text tokens corresponding to its own speech, its inner monologue, which greatly improves the quality of its generation. A small Depth Transformer models inter codebook dependencies for a given time step, while a large, 7B parameter Temporal Transformer models the temporal dependencies.

...It provides both a graphical user interface and API access, making it suitable for casual users as well as developers who want to integrate voice transformation into their own applications. The project also supports GPU acceleration, enabling faster inference and smoother real-time performance on compatible hardware. Additionally, it includes tools for training and managing voice models, giving users the ability to create personalized voice profiles.

...Encodec has applications in speech and music compression, generative modeling, and efficient data transmission for communication systems. The repository includes pretrained checkpoints, PyTorch inference code, and examples for integrating Encodec as a module in downstream generative or streaming systems.

...A pre-trained English model is available for use and can be downloaded following the instructions in the usage docs. If you want to use the pre-trained English model for performing speech-to-text, you can download it (along with other important inference material) from the DeepSpeech releases page.

RNNLIB is a recurrent neural network library for sequence learning problems. Applicable to most types of spatiotemporal data, it has proven particularly effective for speech and handwriting recognition. full installation and usage instructions given at http://sourceforge.net/p/rnnl/wiki/Home/