TRULEO uses a variety of audio analytics and natural language processing models to transcribe and interpret body-worn camera audio.
Speech-to-text Transcription
TRULEO uses a proprietary, internally-built model to transcribe BWC audio to text with industry-leading accuracy. Our model is purpose built in that it focuses on the interaction between the officer wearing the camera and with whom they are conversing with, ignoring other parties that may be much further away from the interaction.
Speaker Diarization
Speaker diarization is the task of a machine deciding "who said what?" TRULEO uses industry-leading speaker diarization that anonymously separates out community member audio from officer audio, then specifically identifies the officer wearing the camera using an Officer ID model.
Officer ID
After TRULEO starts processing a department’s data, our software generates officer voice identification models automatically. In order to obtain the highest level of officer voice print accuracy, Truleo's success team reviews and validates voice prints for each officer within the first 30 days to ensure the highest level of accuracy.
The Officer ID model separates the audio of the officer wearing the camera from other people speaking, including other officers and community members. All of these individuals are kept as anonymized speakers and not identified to maintain privacy.
Translation
In cases where TRULEO's model automatically detects and transcribes Spanish, our translation model automatically translates the Spanish text to English.
Event Detection
TRULEO's Large Language Models (LLM) analyze the transcript and identify dozens of events with very high (99%) accuracy on a segment level. See a full list of event label definitions here.
PII Redaction
TRULEO's LLMs also automatically redact personal identifiable information (PII) to protect community member privacy.
Summarization
TRULEO optionally summarizes a transcript using a Generative AI model into a short body of text that can be used to understand an interaction or write a report.