Vehicle Speech Recognition Model Training via Audio Loopback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vehicle systems face challenges in accurately recognizing voice commands due to variations in speech and ambient sounds, leading to inconsistent actuation of vehicle subsystems, and lack a robust method for training speech recognition models with diverse data.
Innovation Solution
A system that transmits audio samples including speech and ambient sounds to a vehicle computer, where they are broadcast and recorded, and the data is input into a machine learning program for recognition, with unrecognized speech and detected vibrations used to update a training dataset, allowing for retraining of speech recognition models and detection of subsystem variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If speech recognition is performed in vehicle systems with ambient sounds, then voice commands can be recognized, but recognition accuracy deteriorates due to variations in speech and ambient sounds
Solution Approach 1:
The system performs preliminary actions by collecting audio samples from multiple vehicles and using them to train speech recognition models before deployment. Audio samples including various speech patterns and ambient sounds are gathered and used to pre-train models, so that when actual voice commands are received in diverse vehicle environments, the models are already prepared to handle variations in speech and ambient sounds, thereby maintaining high recognition accuracy
Solution Approach 2:
The system implements feedback mechanisms where recognized speech is compared with actual audio samples, and recognition results are fed back to continuously improve the speech recognition models. The system collects feedback from multiple vehicles about recognition performance and uses this feedback to retrain models, thereby improving recognition accuracy in the presence of ambient sounds while maintaining ease of voice command operation
2Measurement precision
If speech recognition models are trained with diverse data from multiple vehicles, then recognition accuracy improves, but system complexity increases
Solution Approach 1:
The system applies universality by creating a centralized training system that serves multiple vehicles simultaneously. A single speech recognition model training platform collects audio samples from multiple vehicles, processes them centrally, and deploys improved models to all vehicles. This multi-functional system achieves diverse data training across the vehicle fleet without requiring complex individual training systems in each vehicle, thereby improving recognition accuracy while managing system complexity through centralization
Solution Approach 2:
The system uses copying by creating replicas of the centralized training system across multiple vehicles. Each vehicle contains a copy of the speech recognition model and related processing capabilities, allowing diverse audio samples from multiple vehicles to be collected and used for training. This copying approach enables distributed data collection with centralized processing logic, improving recognition accuracy through diverse training data while avoiding the need for entirely unique complex systems in each vehicle
3Reliability
If audio samples are collected and processed in real-time, then speech recognition can be improved continuously, but processing time and energy consumption increase
Solution Approach 1:
The system implements periodic action by collecting audio samples continuously but performing model retraining at scheduled intervals rather than in real-time for every sample. Audio samples are accumulated over periods and batch-processed for model training, allowing continuous improvement of speech recognition reliability while avoiding the time and energy costs of real-time processing for each individual sample. This periodic batch processing approach maintains reliable recognition performance without excessive processing delays
Data Source
AI summary
An audio sample including speech and ambient sounds is transmitted to a vehicle computer. Recorded audio is received from the vehicle computer, the recorded audio including the audio sample broadcast by the vehicle computer and recorded by the vehicle computer and recognized speech from the recorded audio. The recognized speech and text of the speech are input to a machine learning program that outputs whether the recognized speech matches the text. When the output from the machine learning program indicates that the recognized speech does not match the text, the recognized speech and the text are included in a training dataset for the machine learning program.