The program was designed to be used in customer services rating; after the service, the customer should talk about its experience. In this case, the audio recorded would be sent to the Speech-To-Text services for transcription, and the transcription would be sent the Natural Language Understanding services to be analyzed in three categories:
- sentiment, an scale between -1 and 1 indicating the level of positivity, where -1 is absolutely negative and is absolutely positive
- emotions, where the level of five emotions (joy, sadness, fear, anger and disgust) would be indicated between 0 and 1;
- keyword spotting, where the most important expressions are indicated with a level of relevance between 0 and 1.
Is mandatory to install the requirements listed in requirements.txt. This can be done using pip with:
pip install -r requirements.txt
Is required to set a dotenv with the credentials for using the Watson Developer Cloud Services. For getting these, you should register on Bluemix, create the Speech-To-Text and Natural Language Understanding APIs and get the credentials. The dotenv file should follow this model:
`BLUEMIX-STT-USERNAME=speech-to-text_user`
`BLUEMIX-STT-PASSWORD=speech-to-text_password`
`BLUEMIX-NLU-USERNAME=natural-language-understanding_user`
`BLUEMIX-NLU-PASSWORD=natural-language-understanding_password`
The program can make the analysis in a role of languages. Therefore, the three categories are supported only in english, the other languages can be used, but will be made an parcial analysis. The default language is US english, but this can be changed by using the flag -language LANGUAGE, where LANGUAGE
Available languages:
Code for the -language flag |
Language | Supported categories |
|---|---|---|
| none (default) | US english | sentiment, keyword and emotion |
| en-UK | UK english | sentiment, keyword and emotion |
| es-ES | spanish | sentiment and keyword spotting |
| fr-FR | french | sentiment and keyword spotting |
| pt-BR | brazilian portuguese | sentiment |
| ar-AR | arabic | sentiment |
The analysis is based in audio files. The audio files can be recorded on running the program or can be uploaded. By default, the audio is recorded when the running begins. It is possible to change this using the flag -in FILE_PATH, where FILE_PATH is the path of the audio file. The allowed audio extensins are wav, flac, ogg, l16, webm, mulaw, basic
When recording an audio, is important to stay in a quiet place, otherwise the audio quality may not be good. The audio is recorded automatically -- the programs detects the silence after the speech and interrupts the recording; then, it is better to avoid long pauses.
The flag -h|--help can be used to show the instructions to change the language or to input a existing file audio.
The results are saved in .csv files in the directory ./results/. They can be analyzed running the program processing.py. As before, the language can be defined with the -language flag, and the default is US english. The analysis shows informations about the ratings made: keywords counting, with relevance rate, and average sentiments and emotions level.
This program was developed by Ana de Paula in may,2017.