Voice Annotation and Data Entry for Revolutionary Speech Recognition Product

Our Client - A Company Leveraging Speech Recognition Technology

The client is a company leveraging the latest speech recognition technology in order to create a system that supported spoken communication. With the number of voice recognition products increasing every day, this client is trying to build a business that helps other implement this technology in an effort to build deeper human connections and connect more communities, making them accessible and inclusive. Connecting businesses through better communication is also another high concern for this client as they work to make voice recognition software more accessible in industries like manufacturing, finance, and retail.

In order to live up to the promises promoted with their new speech recognition tool, the client had to ensure that all information in the system was accurate.

Annotation Requirements Were Set Forth by the Client

The client discovered Data Entry Outsourced (DEO) when looking for potential outsourcing providers for their annotation services and then provided our team with the requirements. They had created a voice recognition tool that was based on a specific voice database so DEO was tasked with the following requirements:

  • Compare the voice output from the voice recognition tool to the files in the client defined database.
  • Annotate (label) each voice sound to show whether there was background noise, another person speaking, or music playing in the background – any additional sounds had to be annotated.
  • Any sounds that were left out of the voice database or the dictionary provided by the client were to be labeled as “not available.”

DEO Faced a Couple of Challenges with This Voice Annotation Project

The biggest challenge here was validating the text/sound output being provided by the tool. Our team had to go through an extensive training course on using the tool and also learn how the speakers pronounce certain words. In fact, the biggest challenge in speech recognition technology in general is the diversity in the way people speak so our team had to face the same challenge that had been plaguing this industry for 70 years.

Step-by-Step Process for Approaching This Annotation Project

  1. Our team discussed the project in detail before we started working on the bulk audio files. We used the trail as a template for any revisions and/or processes that needed to be adapted. Once the client agreed to the terms, we could begin working on the audio files.
  2. We logged into the tool to begin the annotation process.
  3. We grabbed the original audio file and then transcribed its data along with labels into the system.
  4. Our team listened to each of the 20,000 audio files in order to validate all of the data that was transcribed in the previous step. We then labeled each file accordingly.
  5. Once all 20,000 audio files had gone through the annotation and validation process, we submitted them to the client to await feedback.

Our In-Depth Annotation Process Ultimately Benefitted the Client

One of the most difficult obstacles faced by our client was that there were not a lot of annotation providers that could handle such a large workload. We were able to take on the large project and work with them.

  • The client was able to work with a single firm rather than have to outsource to several different providers. This saves them time, money, and ensured high quality results.
  • Labels were 100% accurate so the client did not need to spend any time revising our work. Again, this saved them time and money.
Ask for a Free Trial!