Transcription-free Fine-tuning Of Speech Separation Models For Noisy And Reverberant Multi-speaker Automatic Speech Recognition

Abstract

One solution to automatic speech recognition (ASR) of overlapping speakers is to separate speech and then perform ASR on the separated signals. Commonly, the separator produces artefacts which often degrade ASR performance. Addressing this issue typically requires reference transcriptions to jointly train the separation and ASR networks. This is often not viable for training on real-world in-domain audio where reference transcript information is not always available. This paper proposes a transcription-free method for joint training using only audio signals. The proposed method uses embedding differences of pre-trained ASR encoders as a loss with a proposed modification to permutation invariant training (PIT) called guided PIT (GPIT). The method achieves a 6.4% improvement in word error rate (WER) measures over a signal-level loss and also shows enhancement improvements in perceptual measures such as short-time objective intelligibility (STOI).

Stats

• citations2
• S2 citations—
• github stars0
• HF likes0
• heat score3.58
• arxiv keyravenscroft2024transcription

Related papers

• Directed Speech Separation For Automatic Speech Recognition Of Long Form Conversational Speech (2021)2.26
• Single-channel Multi-talker Speech Recognition With Permutation Invariant Training (2017)12.10
• Separating Long-form Speech With Group-wise Permutation Invariant Training (2021)4.52
• Investigation Of Practical Aspects Of Single Channel Speech Separation For ASR (2021)7.81
• Elevating Robust Multi-talker ASR By Decoupling Speaker Separation And Speech Recognition (2025)0.00
• Permutation Invariant Training Of Deep Models For Speaker-independent Multi-talker Speech Separation (2016)0.00
• Recognizing Multi-talker Speech With Permutation Invariant Training (2017)12.81
• Progressive Joint Modeling In Unsupervised Single-channel Overlapped Speech Recognition (2017)11.67