Hotword-Aware Speech Synthesis to Prevent Unintended Wake-Ups
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Synthesized speech containing hotwords unintentionally activates nearby devices in speech-enabled environments, leading to unintended wake-up processes and power consumption.
Innovation Solution
Implement a hotword-aware model in speech synthesis devices to detect and prevent the initiation of wake-up processes by generating audio output signals that evade detection by nearby device hotword detectors, or training hotword detectors to recognize synthesized speech.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If speech synthesis devices output synthesized speech containing hotwords, then the speech synthesis device can provide comprehensive text-to-speech conversion, but nearby user devices will unintentionally wake up and begin processing
Solution Approach 1:
The patent applies preliminary action by training the hotword detector model in advance with synthesized speech data before actual operation. This pre-training enables the detector to recognize synthesized speech patterns and ignore hotwords within them, preventing unintended wake-ups before they occur. The model is prepared beforehand with knowledge of how synthesized speech differs from human speech, allowing it to filter out false triggers proactively.
Solution Approach 2:
The patent changes the parameters of the hotword detection system by introducing a new parameter: synthesized speech detection capability. The hotword detector model is enhanced to analyze additional acoustic features and patterns that distinguish synthesized speech from human speech. By modifying the detection parameters to include synthesized speech recognition, the system can identify and ignore hotwords within synthesized speech while remaining sensitive to genuine user commands.
2Reliability
If hotword detectors are trained to recognize synthesized speech, then unintended wake-ups can be prevented, but the device complexity increases
Solution Approach 1:
The patent applies universality by designing a hotword detector model that performs multiple functions simultaneously: traditional hotword detection and synthesized speech detection. Instead of creating separate systems for each function, the model is trained to handle both tasks with a single unified architecture. This multi-functional approach improves reliability by preventing false wake-ups while avoiding the complexity overhead of maintaining multiple independent detection systems.
Solution Approach 2:
The patent uses copying by creating training samples that replicate synthesized speech characteristics. The training process involves generating copies of hotwords as they would appear in synthesized speech, allowing the model to learn the patterns without requiring complex real-time analysis. By using synthesized training data that mirrors actual synthesized speech outputs, the model can be trained efficiently to recognize these patterns without increasing operational complexity.
3Measurement precision
If the hotword detector model is trained with synthesized speech data, then it can distinguish between human speech and synthesized speech, but the training process becomes more complex
Solution Approach 1:
The patent applies self-service by having the speech synthesis device generate its own synthesized speech outputs and use these as training data for the hotword detector model. The system trains itself by creating the training samples it needs through its own speech synthesis capabilities. This self-service approach eliminates the need for external datasets or complex data collection processes, making the training process easier while maintaining high detection accuracy.
Solution Approach 2:
The patent merges the speech synthesis function and the hotword detection training process into a unified workflow. The synthesized speech generated for normal operation is simultaneously used as training data for the hotword detector. By combining these two functions, the system avoids the complexity of separate data collection and training pipelines. The merging allows the same speech synthesis engine to serve dual purposes: generating output speech and providing training samples for improved detection accuracy.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
A method (400) includes receiving text input data (302) for conversion into synthesized speech (160) and determining, using a hotword-aware model (320) trained to detect a presence of a hotword (130) assigned to a user device (110), whether a pronunciation of the text input data includes the hotword. The hotword is configured to initiate a wake-up process on the user device for processing the hotword and/or one or more other terms following the hotword in the audio input data. When the pronunciation of the text input data includes the hotword, the method also includes generating an audio output signal (304) from the text input data and providing the audio output signal to an audio output device (118) to output the audio output signal. The audio output signal when captured by an audio capture device of the user device, configured to prevent initiation of the wake-up process on the user device.