Speech Endpoint Detection Using Silent Phoneme Frame Counting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automatic speech recognition (ASR) systems face challenges in efficient decoding and endpoint detection due to high computational requirements and bandwidth usage, with traditional endpointing techniques being imprecise, especially in the presence of background noise and non-speech events.
Innovation Solution
Implementing a hybrid decoding approach using a WFST accelerator with hardware components for local processing of acoustic features and a personalized grammar model, along with an endpointing module to accurately detect speech endpoints by counting silent frames and utilizing a threshold-based method.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional endpointing techniques are used, then the system is simple to implement, but the endpoint detection precision deteriorates in the presence of background noise and non-speech events
Solution Approach 1:
The endpoint detection function is segmented into a dedicated hardware module within the WFST accelerator, separating it from the main processing pipeline. This dedicated module independently counts silent frames and detects endpoints, improving precision without significantly increasing overall system complexity.
Solution Approach 2:
A silent frame counter is introduced as an intermediary component that tracks the number of consecutive silent frames. This counter acts as a mediator between the audio input and endpoint detection logic, providing a quantitative basis for more accurate endpoint detection in noisy environments.
2Productivity
If all speech data is processed remotely via server, then comprehensive processing can be performed, but bandwidth usage and processing time increase
Solution Approach 1:
The endpoint detection function is extracted from the remote server processing and implemented locally within the WFST accelerator hardware. This extraction enables the system to autonomously determine speech endpoints without continuous server communication, significantly reducing bandwidth usage and processing time while maintaining decoding speed.
3Reliability
If the speech recognition system processes all audio data continuously, then no speech is missed, but computational resources are wasted during silent periods
Solution Approach 1:
The system performs preliminary action by counting silent frames before triggering endpoint detection. This preliminary counting mechanism allows the system to anticipate the end of speech segments and pause processing during confirmed silent periods, reducing energy consumption while maintaining reliability through the threshold-based detection approach.
Data Source
AI summary
Embodiments describe a method for speech endpoint detection including receiving identification data for a first state associated with a first frame of speech data from a WFST language model, determining that the first frame of the speech data includes silence data, incrementing a silence counter associated with the first state, copying a value of the silence counter of the first state to a corresponding silence counter field in a second state associated with the first state in an active state list, and determining that the value of the silence counter for the first state is above a silence threshold. The method further includes, determining that an endpoint of the speech has occurred in response to determining that the silence counter is above the silence threshold, and outputting text data representing a plurality of words determined from the speech data that was received prior to the endpoint.