In order to use the Tuner, you must first obtain data in the form of .callsre files generated by the Speech Engine and client applications (see the Engine documentation for information about generating these files).
The .callsre files contain a variety of information about each interaction with the Engine: which grammars were used, what the results of each decode was, various Engine settings, etc.
This page describes how to obtain and import these files manually. We reccomend using the Call Indexer service for this function.
When importing data manually, the simplest option is to just import one or more .callsre files generated by the Engine. To do this, go to File and then select Open Callsre File(s).
You will be prompted to select one or more .callsre files. Once you have done that, details from the calls will populate the Tuner.
Importing Other Data Sources
In addition to loading .callsre files, the Speech Tuner can be used to examine a variety of other data sources. The most common of these will be a Tuner Database file. This is a plaintext comma-separated (CSV) file that includes information about audio, grammars, transcripts, recognition results, and more.
When you save data from the Tuner (e.g. once you have loaded .callsre files and wish to save test results, transcripts, etc.), you will save a Tuner database file.
The format of the file is for each interaction to be on its line, with the following fields separated by commas:
- Transcript Semantic Interpretation (string)
- Transcript text (string)
- Decode Semantic Interpretation (string)
- Decode text (string)
- Interpretation Score (of interpretation index 1) (int)
- Decode time (in milliseconds) (int)
- Audio Filepath (string)
- Acoustic distance (float)
- Result callsre filepath (string)
- Error String (string)
- Model used (string)
- Audio playback time (in milliseconds) (int)
- Word info buffer (binary)
- Speech Quality (enumerated) (int)
- Speech Gender (enumerated) (int)
- Transcript comments (string)
- CSP number (string)
Note that when you save a Tuner file, you will still need to retain the .callsre files associated with the file (as the Tuner database only includes a path to the .callsre in order to get the audio). If you wish to save a single file that includes data and audio, you can choose File and then Save and then pick Save as Tuner Zip. This will produce a zip file that includes all the data needed.
The last format the Tuner understands is a simple transcript file. This is useful when you have audio and transcripts but no .callsre files. For instance, if you were just getting started with LumenVox, you might build a transcript file in order to test how well LumenVox performs for your data.
A transcript file is just another CSV file. Each line contains the transcript, followed by a comma, and then the path to the audio file.