LibriSpeech
PulseAugur coverage of LibriSpeech — every cluster mentioning LibriSpeech across labs, papers, and developer communities, ranked by signal.
10 day(s) with sentiment data
-
OpenAI's Whisper large-v3 available as GGUF package for offline transcription
OpenAI's Whisper large-v3 model has been released as a GGUF package, making it compatible with transcribe.cpp. This native C/C++ runtime, built on GGML, allows for fully offline speech-to-text transcription without requ…
-
Liquid AI releases 74M-parameter streaming audio codec, LiquidCodec
Liquid AI has launched LiquidCodec, a 74 million parameter streaming audio codec designed to efficiently convert speech into tokens for language models and back into audio. The codec boasts impressive speed, achieving o…
-
Earlybird speech model predicts words 90ms before spoken
Researchers have developed Earlybird, a novel speech recognition model that significantly reduces latency by predicting words before they are fully spoken. Unlike traditional streaming models that output words after com…
-
New E-Gaborformer speech model offers smaller memory, lower error rates
Researchers have developed E-Gaborformer, a novel speech recognition model designed for real-time voice agents. This strictly causal encoder utilizes Gabor-dictionary convolutions and a delta-rule memory, resulting in a…
-
New benchmark LibriSpeech-Real-QA targets real-time speech interaction for LLMs
Researchers have introduced LibriSpeech-Real-QA, a new benchmark designed to evaluate speech large language models (SLMs) in real-time streaming scenarios. Current SLMs often process audio offline, waiting for complete …
-
AssemblyAI launches $15/million char TTS API for voice agents
AssemblyAI has launched AssemblyAI-TTS-1, a new text-to-speech API designed for voice agents in live phone calls. The service offers competitive pricing at $15 per million characters and boasts low latency, with a media…
-
New Text Injection Method Speeds Speech Recognition Training
Researchers have developed a new method for training speech recognition models using unpaired text, which speeds up training by 1.8 times compared to previous approaches. This technique, called "Text Injection," replace…
-
Research quantifies ASR errors in TTS evaluation
A new research paper explores the accuracy of text-to-speech (TTS) systems by analyzing the errors introduced by automatic speech recognition (ASR) systems used in their evaluation. The study found that ASR systems miss…
-
Cactus Compute's Whistle speech-to-text model outperforms Whisper on CPU
Cactus Compute has developed Whistle, a compact 16.9 MB speech-to-text model designed to run efficiently on a CPU. Benchmarks on the LibriSpeech test-other dataset show Whistle achieving a 10.47% word error rate, outper…
-
Cactus Whistle speech model challenges Whisper on CPU performance
A new speech-to-text model called Cactus Whistle, developed by Cactus Compute, has been benchmarked against OpenAI's Whisper. This 16.9 MB model runs on a CPU and supports seven languages, offering word timestamps and p…
-
New PhonoBridge system reconstructs silent speech from EMG signals
Researchers have developed PhonoBridge, a novel two-stage framework for reconstructing silent speech from Electromyography (EMG) signals into text. The system utilizes a channel-aware EMG-to-phoneme frontend and a large…
-
AuroraASR-1M: Lightweight Speech Recognition Model Released
AuroraASR-1M, a new speech recognition model, has been released. It features a lightweight design with 949K parameters and a weight-shared Conformer-CTC architecture. The model achieves 11.43% Word Error Rate on the Lib…
-
LoRA adaptation shows mixed gains on Whisper ASR models
A new paper explores the effectiveness of lightweight low-rank adaptation (LoRA) on Whisper-style automatic speech recognition (ASR) models. The research compares zero-shot Whisper, pooled LoRA, and group-specific LoRA …
-
Open-vocabulary speech recognition runs on $5 microcontroller
Researchers have developed Oído, a novel system enabling open-vocabulary English speech recognition to run on a low-cost $5 microcontroller, the ESP32-S3, without requiring a dedicated neural accelerator. Oído achieves …
-
Whistle: Compact Speech Recognition Model Released for On-Device Use
Cactus Compute has released Whistle, a compact speech recognition model designed for on-device applications like mobile phones and robots. This 16.9 MB model runs on a CPU without external dependencies and can perform t…
-
New study explores adapting ASR for child speech while retaining adult performance
Researchers have explored methods for adapting Automatic Speech Recognition (ASR) systems to better understand child speech while retaining performance on adult speech. The study compared techniques like full fine-tunin…
-
DirectSpeech2LLM framework tackles prompt overfitting in Speech-LLMs
Researchers have developed DirectSpeech2LLM, a novel framework designed to address prompt overfitting in Speech-LLMs. This end-to-end system enables LLMs conditioned on speech to maintain instruction-following capabilit…
-
New frameworks enhance LLMs' ability to process and understand speech
Two new research papers propose novel frameworks for integrating speech processing with large language models (LLMs). DirectSpeech2LLM aims to mitigate prompt overfitting in Speech-LLMs by using a CTC loss on frozen LLM…
-
New HearInContext benchmark tests speech recognition's grasp of implicit context
Researchers have introduced HearInContext, a new benchmark designed to evaluate the ability of speech recognition systems to understand implicit context. This benchmark, which includes 3,764 semantic test cases focusing…
-
Pruning Whisper-small improves ASR accuracy by acting as a regularizer
Researchers have demonstrated that neural network pruning can act as a regularization technique for Automatic Speech Recognition (ASR) systems, rather than just a method for compression. By analyzing the sensitivity of …