Acoustic Echo Cancellation Using TTS Data Loopback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Acoustic echo from synthetic speech generated by text-to-speech (TTS) systems interferes with target speech recognition in speech-enabled devices, making it difficult for systems to accurately recognize user inputs.
Innovation Solution
A computer-implemented method using a linear acoustic echo canceller (LAEC) and a neural echo suppressor (NES) to process input audio data streams, determining frame boundaries to align TTS data with echo and target speech, and generating enhanced audio by reducing the echo of synthetic speech.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If synthetic speech is played back through an acoustic speaker, then the speech-enabled device can communicate information to users, but acoustic echo of the synthetic speech is captured by the microphone array, interfering with target speech recognition
Solution Approach 1:
The patent extracts and removes the acoustic echo component from the captured audio signal. The AEC system separates the echo of synthetic speech from the target speech by identifying and eliminating the echo portion, allowing clean target speech to be processed for recognition.
Solution Approach 2:
The patent introduces an intermediary AEC processing system between the microphone array and speech recognizer. This intermediary component processes the captured audio to remove echo before the signal reaches the speech recognition system, preventing echo interference without affecting the speech communication capability.
2Productivity
If the microphone array captures both target speech and echo simultaneously, then the device can maintain continuous listening capability, but the speech recognizer cannot accurately distinguish target speech from echo
Solution Approach 1:
The patent segments the captured audio signal into distinct components: target speech and acoustic echo. By dividing the mixed audio signal into separable parts, the system can process each component appropriately, maintaining continuous capture while enabling accurate recognition of the target speech portion.
Solution Approach 2:
The patent employs feedback mechanisms where the AEC system uses information about the synthetic speech output to generate an echo model, which is then subtracted from the captured audio. This feedback loop continuously refines the echo cancellation process, maintaining both continuous listening and accurate recognition.
3Object-generated harmful factors
If conventional AEC methods are used, then some echo reduction is achieved, but frame boundary misalignment between TTS data and echo causes incomplete echo cancellation
Solution Approach 1:
The patent performs preliminary alignment of frame boundaries between TTS data and captured audio before applying echo cancellation. By pre-synchronizing the temporal boundaries and applying appropriate time delays, the system ensures that the echo subtraction operates on correctly aligned frames, achieving complete echo removal.
Solution Approach 2:
The patent adjusts temporal parameters including frame boundary positions and time delays to achieve precise alignment between TTS data and echo. By dynamically modifying these parameters based on detected boundary positions, the system overcomes the misalignment issue and achieves accurate echo cancellation.
Data Source
AI summary
A method includes receiving text-to-speech (TTS) data and outputting synthetic speech using an audio output device of a user device. The method also includes receiving an input audio data stream captured using an audio capture device of the user device and determining a first frame boundary in the input audio data stream. The input audio data stream includes target speech and an echo of the synthetic speech, while the first frame boundary represents a first alignment of the TTS data and the echo of the synthetic speech. Using a linear acoustic echo canceller, the method also includes determining a second frame boundary in the input audio data stream and processing the input audio data stream based on the second frame boundary to generate enhanced audio. The second frame boundary represents a second alignment of the TTS data and the echo of the synthetic speech.