DTMF Audio Detection and Visualization

One of the more interesting problems of signal processing is how to visualize the contents of a time series signal, especially within an application specific context. The example shown here analyzes a single channel audio sample, looking for DTMF tones, the tones commonly used to encode digits over analog telephone connections.

Signal Detection

The first part of the job is to scan through the audio and to determine what DTMF tones it contains. This is straight forward, and there are many open source and commercial software and hardware systems available for doing so.

Our code -- called dtdect -- specializes on finding DTMF tones in distorted, poorly sampled, or otherwise low quality audio streams. The system locates the tones and marks the start/end of the tones within the audio stream.

Signal Visualization

The second step in the process it the visualization of the search results so that a user can not only see what DTMF tones were found, but can also form an opinion of how accurate the results were. This is done by translating the time/signal nature of the audio sample first into a time/frequency domain which is then mapped into a 2 dimensional visual representation, annotated with the search results as well as certain useful statistics.

The overall result is a simple yet powerful picture of the tones (if any) within the audio sample.