LLM Integration Guide

​

Introduction

Salad Transcription API now offers integration with Large Language Models (LLMs) to provide advanced features such as summarization, translation, custom prompts, and sentiment analysis. By leveraging LLMs, you can gain richer insights and perform complex language processing tasks on your transcriptions.

This guide covers the key LLM-related parameters you can use to enhance your transcription outputs:

• Summarization:
  • summarize
• LLM-Based Translation:
  • llm_translation
  • srt_translation
• Custom Prompts:
  • custom_prompt
• Overall Classification and Sentiment Analysis:
  • overall_classification
  • overall_sentiment_analysis

By properly utilizing these parameters, you can unlock the full potential of LLMs in your transcription workflows.

​

LLM Integration Parameters

​

1. summarize

​

Description

The summarize parameter enables you to generate a concise summary of your transcription using an LLM. You can specify the maximum word count for the summary.

• Default: 0 (No summarization)
• Type: integer

​

Usage

Set "summarize": word_limit in your request to receive a summary with the specified word limit.

Example:

"input": {
  "url": "https://example.com/path/to/file.mp3",
  "summarize": 100
}

Output

The summary will be included in the summary field of the output.

"summary": "This meeting discussed project timelines, budget allocations, and assigned tasks to team members for the next quarter."

​

2. llm_translation

​

Description

Use the "llm_translation" parameter to translate your transcription into one or more specified languages using an LLM.

• Type: string (Comma-separated list of languages)

Usage

Set "llm_translation": "Language1, Language2" to translate the transcription into the specified languages.

Example:

"input": {
  "url": "https://example.com/path/to/file.mp3",
  "llm_translation": "german, italian, french"
}

Output

Translations will be included in the llm_translation object.

"llm_translation": {
  "French": "Votre transcription en français.",
  "German": "Ihre Transkription auf Deutsch."
}

Check translation page for more details.

​

3. srt_translation

​

Description

Translate the generated SRT subtitles into specified languages using an LLM.

• Type: string (Comma-separated list of languages)

Usage

Set "srt_translation": "Language1, Language2" to translate the transcription into the specified languages.

Example:

"input": {
  "url": "https://example.com/path/to/file.mp3",
  "srt_translation": "spanish"
}

Output

Translations will be included in the srt_translation object.

"llm_translation": {
  "Spanish": "1\n00:00:01,000 --> 00:00:04,000\nSu transcripción en español.\n\n..."
}

Check translation page for more details.

​

4. custom_prompt

​

Description

Provide a custom prompt to guide the LLM in performing specific tasks, such as generating a tailored summary, extracting key information, improve result, or answering questions based on the transcription.

• Type: string

Usage

Set "custom_prompt": "Your custom instruction here" to direct the LLM. As a result the LLM model will receive a prompt in the following format: custom instruction:transcription

Example:

"input": {
  "url": "https://example.com/path/to/file.mp3",
  "custom_prompt": "List all action items discussed in the meeting."
}

Output

The LLM will generate a response based on the custom prompt. The result will be included in the llm_result field.

json Copy code

"llm_result": "- Prepare the project proposal by Friday.\n- Schedule a follow-up meeting next Monday.\n- Allocate resources for the development team."

​

5. classification_labels and overall_classification

​

Description

Use the classification_labels parameter in conjunction with overall_classification to classify the entire transcription into specified categories using an LLM.

• classification_labels:
  • Type: string (Comma-separated list of labels)
• overall_classification:
  • Default: false
  • Type: boolean

​

Usage

Set "overall_classification": true and provide your labels in "classification_labels": "Label1, Label2" to classify the entire transcription.

Example:

"input": {
  "url": "https://example.com/path/to/file.mp3",
  "overall_classification": true,
  "classification_labels": "Interview, Meeting, Presentation"
}

Output

The classification result will be included in the overall_classification field.

"overall_classification": "Meeting"

Notes

• Custom Labels:: You can define any categories relevant to your use case.
• Multiple Labels:: The LLM will select the most appropriate label from the list provided

​

6. overall_sentiment_analysis

​

Description

Analyze the overall sentiment of the transcription using an LLM.

• Default: false
• Type: boolean

​

Usage

Set "overall_sentiment_analysis": true to perform sentiment analysis.

Example:

"input": {
  "url": "https://example.com/path/to/file.mp3",
  "overall_sentiment_analysis": true
}

Output

The result will be included in the overall_sentiment field.

json Copy code

"overall_sentiment": "Positive"

​

7. custom_vocabulary

​

Description

Improve transcription accuracy by providing a custom vocabulary of terms that are specific to your domain, such as industry jargon, acronyms, or proper nouns.

• Type: string (Comma-separated list of terms)

​

Usage

Set "custom_vocabulary": "Term1, Term2" to include custom terms in the transcription process.

Example:

"input": {
  "url": "https://example.com/path/to/file.mp3",
  "custom_vocabulary": "SaladCloud, AI Transcription, LLM Integration"
}

Notes

• The custom vocabulary helps the LLM update domain-specific terms.
• Result will have both the original transcrioption and updated under llm_custom_vocabulary.