Speaker Attribution in Distributed Meeting Audio Transcription
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing conferencing tools are inadequate for recording and transcribing ad-hoc meetings, as they require setup and are often unavailable outside conference rooms.
Innovation Solution
A computer-implemented method that processes audio streams from distributed devices to perform speech and speaker recognition, aligning hypotheses to create a best speaker-attributed word transcript.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing conferencing tools with fixed speakers and towers are used, then speaker attribution and transcript generation can be achieved, but the system requires prior setup and is unavailable outside conference rooms
Solution Approach 1:
The system segments the traditional fixed conferencing device into distributed components - multiple mobile devices (smartphones, tablets, laptops) each with its own microphone array. This allows the system to be assembled ad-hoc from distributed segments rather than requiring a pre-configured fixed system, enabling availability outside conference rooms while reducing setup complexity
Solution Approach 2:
The system makes universal use of standard mobile devices that participants already possess, eliminating the need for specialized conferencing hardware. Each mobile device serves multiple functions - audio capture, video capture, and communication - replacing the need for dedicated fixed speakers and towers, thereby enabling ad-hoc meetings without prior setup
2Ease of operation
If distributed devices are used for audio capture, then setup-free operation is achieved, but audio streams from multiple devices must be synchronized and aligned
Solution Approach 1:
The system employs feedback mechanisms where each mobile device transmits metadata about its audio stream (timing information, spatial characteristics) back to the central processing system. This feedback enables automatic synchronization and alignment of audio streams from multiple devices without manual setup, while the centralized processing manages the complexity of coordinating multiple sources
3Productivity
If speech recognition is performed on each audio stream independently, then processing efficiency is improved, but alignment and merging of hypotheses becomes complex
Solution Approach 1:
The system segments the speech recognition process into independent parallel processing of each audio stream, improving productivity by allowing simultaneous processing. The complexity of alignment is then managed separately through a centralized hypothesis alignment module that receives processed results from all segments and integrates them using timing metadata and spatial information
Solution Approach 2:
The system introduces an intermediary hypothesis alignment module that acts as a mediator between the independent speech recognition processes and the final transcript generation. This intermediary layer receives processed hypotheses from all audio streams, performs alignment and merging based on timing and spatial data, and produces the final coordinated transcript, thereby isolating the alignment complexity from the efficient independent processing
Data Source
AI summary
A computer implemented method processes audio streams recorded during a meeting by a plurality of distributed devices. Operations include performing speech recognition on each audio stream by a corresponding speech recognition system to generate utterance-level posterior probabilities as hypotheses for each audio stream, aligning the hypotheses and formatting them as word confusion networks with associated word-level posteriors probabilities, performing speaker recognition on each audio stream by a speaker identification algorithm that generates a stream of speaker-attributed word hypotheses, formatting speaker hypotheses with associated speaker label posterior probabilities and speaker-attributed hypotheses for each audio stream as a speaker confusion network, aligning the word and speaker confusion networks from all audio streams to each other to merge the posterior probabilities and align word and speaker labels, and creating a best speaker-attributed word transcript by selecting the sequence of word and speaker labels with the highest posterior probabilities.