This document describes best practices for recording high-quality audio files for IVR. This is important when you deploy surveys, as the type of voice and quality of the audio recording will affect response rates and data quality.  Below, we outline recommendations to: 

  1. Identify and select a voice and recording partner.

  2. Manage the recording process.  

With some preparation and proactive management, you can easily and quickly record and test audio files, even for long and complex surveys.

Guidelines: Selecting a Partner

Do decide what languages and dialects are important.

Before drafting the IVR audio script, decide which languages and dialects your survey will support, then ensure you have complete questionnaire scripts for the languages you plan to deploy.

Do identify at least 2-3 professional and fluent individuals. 

We recommend working with professional audio talent from a local firm.  Also, be sure to identify 2-3 individuals fluent in each language the survey will use.  

Do collect audio samples from a common script.  

Once you obtain audio samples from a number of individuals, the survey team and fluent speakers should rate which one they like best and why they were good or bad. The selected sample should be easy to understand, trustworthy, and able to record the full script in a timely fashion.

Don't wait until the last minute to find individuals to record audio.  

The process of recording audio content can take some time. We recommend allowing at least 2 weeks to record, test, and finalize audio recordings. These 2 weeks begin after professional talent is secured, and the recording process can begin. If you expect any changes in the script, re-recording audio adds additional complexity. You should identify someone early and begin audio recording iteratively as soon as the questionnaire and languages are determined.

Guidelines: Managing the Recording Process

Do clearly communicate expectation for turnaround times. 

It can take several cycles to arrive at the final set of audio files for an Interactive Voice Response project. Factors such as last-minute changes to the script or poor quality audio require testing of content and resolution of any issues. To fix issues, you must have clear expectations on turnaround time for doing the recordings (e.g., 24-hour or 72-hour turnaround time).

Do test each new version of audio files in the IVR script.

Each time new audio files are loaded, test the new questionnaire to ensure quality.  By taking an iterative approach, questionnaire designers can quickly evolve and ensure high-quality surveys.

Do have audio files versioned and documented against a script.

This step can get complicated and confusing if you are not diligent in naming and versioning the audio files.  Be sure to have identifiers per audio file that relate to the script and can be versioned in the filename (e.g., 1.Consent_v3.wav or 7.Smoking_v1.mp3).

Do have audio files recorded as either.WAV or MP3 files.

Use common file formats for audio files such as .WAV or .MP3. The Ask tool may not accept obscure file formats.

Do consider using online file storage to save versioned audio files.

Managing the versions from your inbox can get complicated quickly. If you store audio files on a local hard drive, the device could suffer loss or damage. Using shared cloud-based storage such as Dropbox or Onedrive can be a great option, particularly in low bandwidth situations.

Do use the same human voice for each language you deploy.

Participants will be confused if they hear different voices in a single survey. This issue can become challenging if someone wants to reuse audio files years later, but small changes are needed. In these instances, if the original voice is no longer accessible, all the audio may need to be re-recorded.

Don't use audio files that have any hiss or crackle noise.

Background noise often can become an issue in IVR projects, distracting users and depressing response rate.  Background noise during recording, poor quality recording equipment, or the post-processing of audio files can all cause these problems.

Don't use text-to-speech for important or large-scale surveys.

We do not recommend using the robo-voice associated with text-to-speech for important or large surveys. It will mispronounce words and depress user participation. Only use the text-to-speech if it is an emergency or you are testing the system using it on a small group of people. 

More Information