Text-Based Echo Cancellation for Speech Devices

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing echo cancelation systems face challenges in efficiently removing synthesized playback audio echoes in speech-enabled devices, leading to compromised speech recognition accuracy and increased latency due to large audio file sizes and resource-intensive processing.

Innovation Solution

A computer-implemented method using a textual echo cancelation system that encodes synthesized playback audio into a text embedding representation, generates cancelation probabilities, and employs a cancelation neural network to remove echo from overlapped audio signals, leveraging a text-to-speech system's existing text processing to reduce latency and resource usage.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Object-affected harmful factors

If traditional acoustic echo cancellation methods are used to remove synthesized playback audio echoes, then echo reduction is achieved, but processing latency increases and computational resources are heavily consumed

Engineering Contradiction:
ImproveechoVSAvoidprocessing latency
Core Design Contradiction:
Object-affected harmful factorsVSLoss of time

Solution Approach 1:

The patent extracts only the essential information needed for echo cancellation (textual representation of synthesized playback audio) from the complete audio signal, rather than processing the entire audio file. This extraction approach allows the system to identify and cancel echoes by comparing the overlapped audio signal with the textual representation, significantly reducing processing latency while maintaining effective echo reduction

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent replaces traditional acoustic echo cancellation mechanisms (which process audio signals directly) with a textual processing approach. By converting synthesized playback audio into textual representation and using text-based comparison and identification, the system substitutes heavy audio signal processing with more efficient text processing, reducing computational resource consumption and processing time

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Object-affected harmful factors

If traditional acoustic echo cancellation methods are used to remove synthesized playback audio echoes, then echo reduction is achieved, but computational resources are heavily consumed

Engineering Contradiction:
ImproveechoVSAvoidcomputational resources
Core Design Contradiction:
Object-affected harmful factorsVSUse of energy by moving object

Solution Approach 1:

The system extracts only the textual representation of the synthesized playback audio, which is much smaller and more efficient to process than the complete audio signal. This extraction allows the echo cancellation to be performed using text-based operations rather than computationally intensive audio signal processing, significantly reducing computational resource consumption

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent substitutes traditional audio-based echo cancellation mechanisms with a text-based processing system. By replacing audio signal analysis and manipulation with text processing operations, the system achieves echo cancellation with significantly lower computational resource requirements and energy consumption

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Measurement precision

If synthesized playback audio is processed using large audio files, then accurate echo identification is achieved, but file size and processing overhead increase

Engineering Contradiction:
Improveecho identification accuracyVSAvoidfile size
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent extracts the essential informational content of the synthesized playback audio in the form of textual representation, discarding the redundant audio signal data. This extraction maintains the ability to accurately identify echoes (since the text preserves the semantic content) while dramatically reducing file size and processing overhead

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

Instead of using the audio signal itself to identify and cancel echoes, the patent inverts the approach by using the textual representation of the audio to achieve the same goal. This inversion allows the system to work with much smaller data structures while maintaining echo identification accuracy

Inventive Principle:
Principle #13The other way round (Inversion)

Data Source

PatentEP4139920B1Text-based echo cancellation
Publication Date: 2024.03.06 GOOGLE LLC
  • EP4139920B1 patent drawingFigure 1
  • EP4139920B1 patent drawingFigure 2A
  • EP4139920B1 patent drawingFigure 2B

AI summary

A method (400) includes receiving an overlapped audio signal (202) that includes audio spoken by a speaker (10) that overlaps a segment (156) of synthesized playback audio (154). The method also includes encoding a sequence of characters that correspond to the synthesized playback audio into a text embedding representation (212). For each character in the sequence of characters, the method also includes generating a respective cancelation probability (222) using the text embedding representation. The cancelation probability indicates a likelihood that the corresponding character is associated with the segment of the synthesized playback audio overlapped by the audio spoken by the speaker in the overlapped audio signal.