Help Center

Getting Started

Welcome to the VoiceServant Platform User Guide. VoiceServant uses state-of-the-art artificial intelligence to turn unstructured audio recordings (such as meetings, interviews, support calls, and voice notes) into structured, clean tabular data.

How it Works

The typical extraction lifecycle on the VoiceServant platform consists of three main steps:

1. Create a Template: Define the blueprint or schema of the output data you wish to extract. You can configure individual objects (like sheets in an Excel document) and fields (with instructions, types, and descriptions).
2. Run an Extraction: Upload your audio files (MP3, WAV, and other common formats) and select the corresponding template. VoiceServant transcribes and processes the recordings asynchronously to extract information exactly matching your schema.
3. Export Data: Review the extracted results in the dashboard. Check for accuracy and export the data directly as an Excel workbook (`.xlsx`) or a CSV file.

Templates

A Template is a user-defined blueprint that specifies the structure of the data you want VoiceServant to extract from your recordings. By defining objects and fields, you instruct the AI on exactly what information to find and how to format it.

Core Template Configuration

When configuring a template, you can customize how extractions are handled:

• Output Mode: Choose between Combined (default) or Single. Combined merges all recording outputs into a single consolidated sheet. Single generates separate outputs for each recording.
• Store Data: Toggle this on to retain the original recordings for audio playback and review. Turn it off to automatically purge files from the server after extraction completes (ideal for strict data privacy).
• Global Instructions: Provide overall context to the AI model, such as the recording language, expected speakers, or specific terminology and acronyms used in the conversation.

Template Creation Process In-Depth

Manual Template Setup

Creating a template manually gives you maximum control. First, add an Object (e.g. `Meeting Summary` or `Action Items`). Within each object, add Fields:

• Field Name: A short descriptor (e.g. `Speaker`, `Decision`, `Due Date`).
• Field Type: Let the AI know the data type. E.g. a date type will enforce ISO-8601 formatting, and a number type will strip surrounding text.
• Field Description / Instruction: Direct prompt telling the AI where to find the data or how to interpret it. For example: "Extract the name of the person who committed to the action item."

XLS Template Upload Setup

If you already have a reporting template in Excel, uploading it is the fastest way to get started. VoiceServant reads the workbook structure:

Once uploaded, you can refine the generated prompt instructions and adjust data types to achieve optimal extraction accuracy.

Run Extraction

Once you have created a template, you are ready to run extractions. VoiceServant transcribes and processes your recordings asynchronously and formats the output data according to your template schema.

Submitting an Extraction Job

To submit a recording for extraction using the web interface:

1. Navigate to the Runs or Extraction section in the dashboard sidebar.
2. Choose the template you want to use from the dropdown selection list.
3. Click the upload zone or drag and drop your audio files. You can upload multiple files to process them in parallel.
4. Configure any overrides (like forcing a specific output_mode or disabling data storage).
5. Click Run Extraction to submit the job.

Supported File Formats & Guidelines

VoiceServant supports a wide variety of audio formats for extraction. For best results, follow these quality guidelines:

Monitoring and Exporting Results

After submission, your extraction job appears in the Run History table. You can watch progress in real time as the status shifts from pending or processing to completed or failed.

Once a run is complete:

• Clicking the run details displays an interactive table of the extracted data alongside the original recording. You can play back the audio while reviewing the extracted results.
• Use the Export button at the top-right of the run page to download results.

Reference Data

Reference Data allows you to upload custom datasets (such as product catalogs, customer rosters, team directories, or internal code tables) and use them to validate, match, and enrich the data extracted from your recordings.

What is Reference Data?

In many workflows, transcribing speech directly isn't enough. You may need to verify that a mentioned customer matches a database record, or map an informal product name spoken on a call to an official SKU. Reference Data bridges the gap by letting you query your uploaded records directly within your extraction prompts.

Creating and Managing Datasets

To upload your lookup data:

1. Navigate to Reference Data in the dashboard sidebar.
2. Click New Dataset to open the creation wizard.
3. Define the basic schema fields (e.g. `SKU`, `Product Name`, `Unit Price`).
4. Upload your records via a CSV import, or manually add individual rows directly within the interface.

Querying Datasets with the Query Builder

Once a dataset is active, you can bind it to a field in a template. This allows you to write lookup queries using the Query Builder:

Lookup Resolution & Return Field

When a query matches one or more records in your dataset, you specify a Return Field (e.g. `SKU` or `Internal Code`). VoiceServant will replace the raw extracted value in the final spreadsheet or JSON output with the matched record's return field. This eliminates manual catalog lookup and post-processing steps.

Best Practices

Optimize your extraction success, increase accuracy, and manage workspace security by applying these guidelines when designing templates and running jobs.

1. Field Instruction Formatting (Prompt Engineering)

The AI reads your field instructions as direct orders. Clear, explicit prompts lead to significantly higher accuracy:

2. Handling Multi-Object Templates

When dealing with complex recordings that contain repeated items or distinct topics:

• Separate Objects: Define a distinct template object for each unique collection. For instance, put general meeting metadata (date, attendees) in one object, and the list of action items in another.
• Consistent Row Definitions: For list-style objects, keep fields simple (e.g. `Owner`, `Task`, `Due Date`). Let the AI do the heavy lifting of mapping each item.

3. Data Privacy and Security

If your organization handles sensitive information (e.g. personal data, confidential financials, or recorded customer conversations):

4. Common Troubleshooting Tips