Speech to text (Arabic) - Based on Microsoft Azure
Budget: $750 – $1,500 USD
This is a proof of concept project in which we need to demonestrate the capabiliteis of Microsoft Azure Speech to Text in Arabic Language. The project will provide 2 interface (Web and Mobile App iOS & Android), in which it will provide an interface for the end user (a judge) where he will dictate the final judgement in Arabic language (Qatari dialect), the system will capture the voice and turn it to text. The system must provide easy to use interface with ability to easily implement corrections. Also the solution must provide APIs that wraps Azure APIs and extent it with the special capabilities we need. Following are the main features of the system:
1. Online and offline voice recognition: The user can either start dictation right away to the system and the text will start to appear as he is speaking. The system should allow the user to upload offline recorded audio file in the known standard format e.g., MP3, AAC, OGG with any bit rate and without restrictions on file size or audio duration.
2. Audio enhancement: during the online dictation, the system must warn the user if his voice is law, and in the background (both online and offline) audio quality enhancement must be made, such as noise cancellation and increasing the volume.
3. Interfaces controls: at least the system must allow the user to pause recording, finish the recording, listen to what has been recorded, and select part of the audio and re-record it.
4.Review interface: the system should provide a review interface in which the user can see all the resulted text. The user can select any part of the text and listen to the corresponding part of the audio clip and either re-record it or type in any required modifications.
5.Predefined terms: the system should allow the user to define certain words and the resulted text, it can be kind of shortcut for lengthy term. The system should allow the user to define certain numbering schemas, e.g., when saying (قضية رقم 394 على 22 من سنة 2021) should be recognized as (قضية 394\22\2021). Also, the user can define how number and months are to be recognized. e.g., when saying (عام الف وتسعة مئة وستة وتسعين) can be recognized numerically as (عام 1996) or as words (عام الف وتسعة مئة وستة وتسعين). Same applied to months names, e.g., when saying (تمت القضية في شهر 3) can be recognized the same or as (تم القضية في شهر مارس).
6.Machine learning: For the common mistakes in recognition that the user usually fixes manually, the system must have some capabilities of enhancing the reorganization to automatically add this to its dictionary or being to learn from the user fixes and fix it on the fly.
7.Languages and dialects: the user should be able to select the dialect of the speaker e.g., Qatari dialect or Formal classical Arabic, also the user during the dictation should have a way to tell the system that he will say something in English and the system should recognize it as English text then he can get back to the normal Arabic recognition.
1. Online and offline voice recognition: The user can either start dictation right away to the system and the text will start to appear as he is speaking. The system should allow the user to upload offline recorded audio file in the known standard format e.g., MP3, AAC, OGG with any bit rate and without restrictions on file size or audio duration.
2. Audio enhancement: during the online dictation, the system must warn the user if his voice is law, and in the background (both online and offline) audio quality enhancement must be made, such as noise cancellation and increasing the volume.
3. Interfaces controls: at least the system must allow the user to pause recording, finish the recording, listen to what has been recorded, and select part of the audio and re-record it.
4.Review interface: the system should provide a review interface in which the user can see all the resulted text. The user can select any part of the text and listen to the corresponding part of the audio clip and either re-record it or type in any required modifications.
5.Predefined terms: the system should allow the user to define certain words and the resulted text, it can be kind of shortcut for lengthy term. The system should allow the user to define certain numbering schemas, e.g., when saying (قضية رقم 394 على 22 من سنة 2021) should be recognized as (قضية 394\22\2021). Also, the user can define how number and months are to be recognized. e.g., when saying (عام الف وتسعة مئة وستة وتسعين) can be recognized numerically as (عام 1996) or as words (عام الف وتسعة مئة وستة وتسعين). Same applied to months names, e.g., when saying (تمت القضية في شهر 3) can be recognized the same or as (تم القضية في شهر مارس).
6.Machine learning: For the common mistakes in recognition that the user usually fixes manually, the system must have some capabilities of enhancing the reorganization to automatically add this to its dictionary or being to learn from the user fixes and fix it on the fly.
7.Languages and dialects: the user should be able to select the dialect of the speaker e.g., Qatari dialect or Formal classical Arabic, also the user during the dictation should have a way to tell the system that he will say something in English and the system should recognize it as English text then he can get back to the normal Arabic recognition.