Speech Analysis Tasks

This document provides a detailed overview of the entire workflow for speech analysis tasks, including creation, configuration, execution monitoring, and result analysis. The system supports two data sources: CDR-based (Call Detail Records) and offline recordings. It offers precise task scheduling mechanisms, visual progress feedback, and in-depth transcription details to help administrators efficiently conduct speech analysis and business reviews.

Overview

Speech analysis tasks are the core vehicle for executing voice analysis. Administrators can flexibly specify the scope of recordings to be analyzed (by automatically pulling CDRs or manually uploading files) and configure execution strategies (immediate or scheduled). After execution, the system provides real-time feedback on success/failure progress and offers multi-level result viewing capabilities, ranging from list overviews to timeline analysis of individual calls, enabling deep mining of call content.

Access

1. Log in to the backend management system.
2. In the left navigation bar, select Speech Analysis > Analysis Tasks to enter the task management page.

Features

1. View Task List

In the speech analysis task list, you can quickly search for tasks using the top filter bar (Creation Time, Task Name, Status, Data Source). The definitions of core fields in the list are as follows:

• Task Name: A custom identifier for the task.
• Data Source: Distinguishes whether the data source is "CDR" or "Offline Recording".
• Time Range: The time coverage of the recordings for the task.
• Execution Method: Displays as "Run Immediately" or "Scheduled Execution".
• Status: Reflects the current lifecycle stage of the task.
  • Pending: The task has been created but has not started yet (requires manual click on "Start").
  • Running: The task is currently processing data.
  • Completed: All data processing is finished.
  • Failed: The task was interrupted due to an exception during execution.
• Progress: Displayed in the format of Success Count/Failure Count/Total Count. For example, 0/3/3 means a total of 3 recordings, 0 successful, and 3 failed; 0/1/1 means 1 recording failed to process.
• Creation Time: The specific time when the task was generated.

2. Create Speech Analysis Task

1. In the upper left corner of the task list page, click + Create Task.
2. In the pop-up dialog box, fill in the basic information:
   • Task Name: Enter an easily recognizable name.
   • Language: Select the language corresponding to the recording (e.g., "Indonesian").
   • Execution Method:
     • Run Immediately: After creation, the task enters a "Pending" status. You need to manually click "Start" on the list page to begin analysis (to allow for modification and confirmation).
     • Scheduled Execution: Set a future trigger time, and the system will automatically start the task on schedule.
3. Select the Data Source according to business needs:
   • Scenario 1: Select "CDR"
     • CDR Time Range: Set the start and end times. The system will automatically pull all call records within this time range as analysis objects.
   • Scenario 2: Select "Offline Recording"
     • Upload Recording: Supports manual uploading of local recording files (formats supported: .wav, .mp3), suitable for model testing or specific case reviews.

3. View Execution Results

For tasks with a status of "Completed" or containing partial successful data, click the Results button in the action column to pop up the recording processing list for that task.

Result List Field Descriptions:

• Callee/Extension Number: Number information of both parties in the call.
• File Name: The name of the original recording file.
• Recording Duration: The duration of the call in seconds.
• Source: Distinguishes the data source (e.g., "Automatic" pull or "Manual" upload).
• Recognition Type: Displays the technical type of speech recognition (e.g., "Single Party Recognized", "Normal Two-way Dialogue").
• Call Tag: Business tags matched by the system based on keywords (e.g., "Normal Call", "Prospective Customer", "Callback Customer").
• Content: A text summary preview of the transcription.
• Recording File: Click Play to directly listen to the original recording.

4. Analysis Detail Record

Click Details in the result list to enter the in-depth analysis page for a single call. This page restores the entire call process and provides refined timeline analysis.

Core Information Area:

• Basic Data: Displays metadata such as callee number, recording duration, source, recognition time consumption, and creation time.
• Call Tag: Highlights the business classification tag matched by the call (e.g., "Prospective Customer").
• Matched Keywords: Lists the specific words that triggered the tag (e.g., "tertarik" in the screenshot).

Dialogue Timeline Analysis:

The bottom of the page displays the dialogue content sentence by sentence in a timeline format, supporting refined review:

• Role Identification: Clearly identifies the speaker as "Agent" or "Customer".
• Silence Detection: Marks the silence duration (e.g., 1.0s) and voice duration (e.g., 2ms) next to each sentence to help analyze pauses and reaction times in the dialogue.
• Keyword Highlighting: Matched keywords are highlighted in the transcribed text for quick location of key information.