This project includes two parts:
- Text to Speech Conversion
- Speech to Text Conversion
However, to help transcribing voice recording to text easier, the main propose of this project is generating a well-organized file that includes each sentence of yoru voice recording files.
- Python 3
- Speech API token of Microsoft Project Oxford
- Bing Voice Recognition token
Save your API and tokens in tokens_sample.py and rename it to tokens.py
In this part, we use Speech API of Microsoft Project Oxford to synthesize the voice from text inputs. You can subscribe a free plan of Speech APIs, which includes 5000 free API calls per month.
Usage:
python synthesizer.py "Never gonna give you up, never gonna let you down."
Output: synthesized.wav
In this part, we should get Bing Voice Recognition token, like Speech API, you can call 5000 times for free per month. Then you can access your keys from this page
Since Bing Voice Recognition service can't recognition our voice recoding file at once, we should separate the file to mutiple parts first. We suggest the tool Audacity for track editing.
- Open your file with Audicity and choose "Analyze → Silence Finder..."
- Change the settings, then press "OK"
- Click "File → Export Multiple..."
- Check export format and name files as the following settings, then export all files to same directory
Put all .wav files in same directory, then use python recognizer.py FOLDERNAME to generate results. Yes, it will try to recognize each sentense in your voice files.
The output of recognizer.py is a .csv file which includes all sentences in your voice recoding files. Each row represent the corresponding output of the voice recording. If the recoginition failed, the output file also represents the reason and error code.
MIT License