Interview AI

One of the most time-consuming steps after conducting an interview is to go over the recording for insights. This is often taken care of manually where the interview is listened to for transcription and note taking, sometimes more than once. Relevance AI provides you with a workflow that takes care of tasks such as

• transcription
• speaker diarization
• insight extraction

How to use Interview AI

All you need to do is to go through the three steps below.

1. Have your data ready
2. Analyse your data
3. See the results

📘

Audio/Video is recommended

Even though this workflow accepts PDF or CSV data as the input, it is highly recommended to use the original audio/Video file(s).

Data

Sing into Relevance AI and on the very first page, click on "Interview AI". This will open up a flow that covers all the steps from 1. upload to 2. analysis and finally 3. interview AI dashboard.

📘

What if I have already uploaded my source files as a dataset to Relevance AI?

This means you have gone through the below two steps:

1. Save your interview audio file(s) in conventional audio formats such as mp4 and upload it to Relevance AI.
2. Go to workflows and open the "Interview AI" workflow wizard. If it is not listed on the Overview page (i.e. under "Suggested workflows for this dataset"), one easy way to access a workflow is to search for it under Browse Workflows as shown in the image below.

2. Follow the workflow wizard:

• Step 1: Name your analysis. This is only required if you wish to run this workflow on the selected dataset multiple times with different setup. Enter a name under which the results will be saved to avoid overwriting an existing analysis result.
• Step 2: Specify your data format. You can use your audio file or the transcription in PDF or text (CSV) format. Depending on the selected option in step two, the rest of the page might be different from the shown image below. This image shows the setup for audio input. However, all other options are included in the following explanation.

• Step 3: Which field contains the data? For Audio or PDF files, specify the field containing the URL to the source file. For text data (i.e. CSV as a source file), specify the field containing the transcription.
• Step 4:
  • For Audio files, specify your preferred model for processing the source audio file(s). These models are different in their speed, level of accuracy and cost. For English data, Relevance AI recommends the Fast model as it is very accurate and not very costly. However, you can use the Premium option for super-high accuracy but keep in mind that it will cost more credits. For non-English data, use Whisper which is the model trained by OpenAI.
  • For PDF data, specify if the pages are scanned document or simply a PDF of a text file. Only select the OCR option, if pages in your PDF are scanned images.
  • For CSV text files, see step 6 below.

• Step 5:

  • For audio: specify the language spoken in the interview. If there are multiple languages, specify the dominant one.
  • For PDF: see step 6 below
  • For CSV text files, see step 7 below.

• Step 6: Length of quotes. Specify the best chunk size. Chunking sizes specify the size of blocks of text to be used when looking for answers and retrieving the best matching quotes. So, for example, if your interview contains short sentences as sources of truth, chunking/splitting by sentences works fine. But if a paragraph is needed to provide a meaningful answer, use the larger options.

• Step 7: Theme extraction method. Relevance AI extract the main and most important quotes from your interview. You can

  1. leave this on AI
  2. specify an objective which makes AI to analyse the interview from a specific angle. An objective can be formed as a simple question. For example "What is the effect of rate rises on people's lives?"
  3. provide your code-frame/taxonomy, so that the extracted codes are labelled with codes where possible. More details are available at Taxonomy Builder. Also read about How To Improve Tag Accuracy and Granularity.

• Optional settings:

  • Receive an email notification upon the workflow completion
  • Adding advance filters (Not recommended for this workflow)
  • Apply to the whole dataset default setting: all new entries (i.e. data sources that have not been included in the analysis before) will be included in this run
  • You can specify a name for the RESULTING dataset.
  • You can specify what is the objective of this analysis. Note that it is as simple as entering what the goal is in natural language. This helps AI to look at the data from a specific angle.

4. Click on "Run workflow"

When the workflow is finalized, under history you can access the specific dashboard that is designed for Interview AI.

📘

The Quote Dataset

After Interview AI finalizes successfully, a new dataset gets added to your account with its name ending in "quotes".

This dataset is composed of important quotes extracted from the interview as well as the full transcription in separate fields. If you wish to further process your data, we recommend using the new dataset.

Note that you might need to refresh the page for the quote dataset to appear on the page.

Interview AI's specific dashboard

By clicking on "View interview AI dashboard", you will be directed to a new dashboard where all your individual source files can be seen separately.

When a source file is selected, you will see all the automatically extracted (or manually entered if the taxonomy was provided in the workflow wizard) topics with the number of relevant quotes found in the transcription.

Scrolling down on the page, the whole transcription and the highlights can be viewed.

The quote dataset and further analysis of the extracted quotes

After Interview AI is finalized, you will have a new dataset whose name ends in _quote. Note that you might need to refresh the page for the dataset to get listed under your account. This dataset is composed of all the important quotes extracted from the interview transcription. The text is ready to be further analyse under AI Tagging or AI Tagging.

How to export the Transcription?

After the workflow is finalized the transcription is added to your original dataset (i.e. even though the quote dataset includes the transcription, it is best to use the original dataset to export the transcription). There are different ways to export the transcribes audio/video.

Export Utterances workflow

📘

Only for Audio/Video Source files

As was noted above, Interview AI works with variety of input data types (e.g. audio, pdf). Exporting utterances per line in a CSV is only available if the source audio/video is directly transcribed by Relevance AI.

This workflow exports the utterances along with selected metadata (e.g. speaker) to a CSV file. Locate the "Export Utterances" under the workflow page as shown in the image below.

Next, select what fields to include in the exported CSV:

• Text: transcription
• Speaker: labels starting from A
• Start/End: denoting the beginning and end time of each utterance
• Filename: original audio/video file name

After the workflow is finalized, you will receive an email with a link to your CSV. You can also download the CSV from the corresponding entry in the History page.

Export the whole transcription and Summaries

One very straightforward way is to use the export option on the Data Table view as shown in the image below. The Export button is located on the top-right.

Keep in mind that you need to specify which fields to be included in the Export. This is possible by clicking on the form and in the Fields to include in Export section.

URL Transcribed is the field in which the full transcription is saved. However, there are other options such as interview questions or interview summary that can be set to export.