Tag
A video demonstrates a GUI front end experiment that uses the Skred sound engine as an audio effects processor, with a design inspired by the Korg DW-8000 digital delay line. The project is MIT-licensed open source on GitHub.
A GitHub project that uses FFmpeg to simulate cassette tape audio profiles including tape noise, wow/flutter, bandwidth limits, and equalizer adjustments. Provides scripts for different cassette types.
Microsoft released vibevoice, a 7B model that transcribes up to an hour of audio in one shot with built-in speaker diarization and timestamps, supporting 50+ languages and running locally without API costs.
MOSS-TD, a speaker-aware ASR system, is optimized within the SGLang-Omni serving stack, enabling 38-minute multi-speaker audio to be transcribed in about 49 seconds on a single H100, with concurrent processing of 16 meetings.
Inkling, a 975B Mixture-of-Experts model (41B active) with 1M context and Apache-2.0 license, introduces a novel audio front end using a 7.9M parameter lookup table instead of a traditional encoder, achieving strong performance on speech tasks. The model was pretrained on 45 trillion tokens of text, images, audio, and video.
A browser-based tool that converts singing voice into printable musical notation, with adjustable parameters for pitch tolerance, noise gate, and note splitting sensitivity.
Mirelo AI, in collaboration with Kyutai Labs, introduces an open-source Audio-to-MIDI model that transcribes full music mixes into separate MIDI tracks per instrument, detecting chords, key, and tempo directly from the mix without requiring isolated stems.
MuScriptor is an open-source model for multi-instrument music transcription, capable of transcribing recordings into MIDI notes for each instrument without prior knowledge of the instruments present.
UniSE is a unified, prompt-free, autoregressive speech enhancement model based on a decoder-only language model, supporting multiple tasks like speech restoration, target speaker extraction, and speech separation in a single model.
A technical blog post describing the development of a real-time system that follows a drummer by estimating tempo and phase from drum hits using a phase-locked loop, enabling the software to adapt to human feel instead of forcing the drummer to follow a machine.
A study from OrukLabs shows that speech models trained solely on transcription or masked audio tasks spontaneously learn to represent emotions in their deeper layers, as revealed by mapping with real voice clips.
audio.cpp releases a major update adding music/SFX generation and source separation with ACE-Step, HeartMuLa, Stable Audio 3, and HTDemucs, achieving up to 10x real-time speed for long music generation in native C++/GGML.
DSPi is an open-source firmware that turns a Raspberry Pi Pico into a fully featured digital audio processor, offering room correction, parametric EQ, matrix mixing, and more.
This paper introduces the first public multimodal dataset of 100 Turkish scam and benign phone calls, evaluating seven LLMs under raw audio, ASR transcripts, and human-corrected transcripts. Results show transcript-based inputs outperform direct audio, highlighting the need for inclusive AI safety research in low-resource languages.
This article shares hard-won lessons from building real-time voice AI agents, highlighting the importance of proper turn-taking, VAD handling, billing awareness, and avoiding echo loops.
A local CLI tool that uses OpenAI's Whisper to detect and remove filler words (um, uh, erm) from audio recordings, employing techniques to avoid audio artifacts like clicks and background hiss.
Hush is an open-source tool for noise suppression designed for voice AI agents, improving audio clarity in real-time interactions.
Microsoft released VibeVoice, an open-source model that processes a full hour of audio in one pass and returns a structured transcript with speaker identification and timestamps, disrupting paid transcription services.
Resonate is a low-latency, low-memory algorithm for perceptually relevant spectral analysis of audio signals, using resonator models with exponentially weighted moving averages.
Santiago highlights the limitation of traditional STT pipelines that lose tone and emotion, then introduces Velma, a voice-native AI model from Modulate that analyzes raw audio to capture intent, emotion, and other acoustic signals, available via API at 10x cheaper than LLM-based approaches.