Gradient-Based Speech-to-Text Alignment for Any ASR Model: From CTC to Speech LLMs
This work provides a universal alignment technique for ASR models, addressing the need for word-level timings in models that lack native alignment, though it is computationally expensive.
The authors propose a gradient-based method for speech-to-text alignment that works for any differentiable ASR model, including those without native alignment like attention-based encoder-decoders and speech LLMs. The method achieves usable alignment across 16 models, often outperforming native alignment in streaming models, but requires one backward pass per token.
Speech-to-text alignment means finding the temporal boundaries of each word in the audio. Some models provide such an alignment directly and others do not. Connectionist temporal classification (CTC) and transducer models have an alignment by construction, whereas attention-based encoder-decoders (AED) and speech large language models (LLMs) do not, and their word timings are usually read off the attention weights instead. All of these signals live on the encoder frame grid, which bounds their temporal precision. We study a generic gradient-based alignment that applies to any differentiable ASR model. We take the gradient of each teacher-forced token log probability with respect to the input, reduce it to a per-frame saliency, and decode the resulting matrix into word boundaries with a single dynamic-programming pass. The method needs no training, no model modification and no alignment heads, works across all model families including the speech LLMs, and aligns on the input grid rather than on the coarser encoder grid. We evaluate it on sixteen models from four families, on read (TIMIT) and spontaneous (Buckeye) speech, each against the model's own native or attention-based alignment. We find that the gradient yields a usable alignment for every model, that it is usually somewhat behind a strong native aligner but better where the native alignment is weak, as for the streaming models, and that its main disadvantage is the cost of one backward pass per token.