Voice Data Import Guidelines:

Adhere to the following specifications for a successful import of voice data into Sprinklr:

File Directory:

• Maintain a single, consistent directory where Sprinklr configuration is searched for files.

• Ensure a one-to-one mapping between a specific importer configuration and a directory. No two importer configurations can point to the same directory.

File Extension and Content:

• Sprinklr currently supports only zip or tar.gz files for voice data.

• Each zip file should contain one metadata file and one audio call recording.

• The CSV in the zip file should have exactly two lines/rows: the first for headers and the second for values.

​

File Prefix:

​

• Maintain a consistent prefix for zip files. The configuration searches for zip files with this consistent prefix (For example, Partner_Call_1.zip, Partner_Call_2.zip).

• Inside zip/tar, if Sprinklr does not find any metadata or recording file with expected extensions, the entire zip/tar file gets ignored.

​

Metadata File Extension:

​

• Sprinklr currently supports three types of metadata file extensions: JSON, XML, and CSV. Partners must adhere to these extensions for all selected files.

​

Metadata Contents:

​

• Every conversation must contain a unique conversation ID for system identification.

• Ensure consistency in metadata contents across all cases for a single configuration.

• In the absence of an agent ID in the config, the system assigns the case to the default bot user shared across partners.

• The system does not currently support the import of customer profiles. If the customer ID is present, a new profile is created; otherwise, the case is associated with a backend-created profile consistent across partners.

• Only one value of an agent associated with a single case is currently supported.

• If the call start time is not present in the metadata, the current time of upload is considered the call start time.

• For CSV metadata files, it should have exactly two lines/rows: the first for headers and the second for values.

Audio Files:

• Supported formats: WAV, MP3, OPUS with a frequency of 8000 Hz or more, and stereo.

• PCM Formats (Choose one):

  • Pcm_s16le

  • PCM A-law

  • PCM mu-law

  • G.729a

• Optimizing for ASR Models:

  • Ensure audio files are representative and mimic actual production data and use cases for optimal ASR (Automatic Speech Recognition) model performance.

  • Include recordings with different speakers, providing a good distribution for both male and female genders.

  • Include sample calls/audios for each intent/use case.

  • Ensure availability of different speakers from various locations or with different accents.

Processing Time:

• On average, it takes around five seconds to create a case with a five minute conversation between a customer and an agent via SFTP.

Sample Metadata File:

​​