When working with voice transcription in Genesys Cloud, keep in mind the following limitations:

• Transcription of ACD agent consults recordings is not supported. For more information, see Record ACD agent consult calls. 
• When an agent initiates a consult or conference with another agent, or directly transfers a call to them, the initiating agent’s queue’s voice transcription settings remain active throughout the interaction. For example, if Agent 1 from Queue A, which has transcription enabled, consults or transfers the call to Agent 2 from Queue B, where transcription is disabled, the entire interaction will still be transcribed. To avoid additional transcription costs, consider using a blind transfer to Agent 2’s queue instead. This allows the receiving agent’s default settings to take effect.
• When an agent initiates a consult with another agent the initiating agent’s queue’s voice transcription settings remain active throughout the interaction and all users involved in the interaction will be considered as transcription users. For example, if Agent 1 from Queue A, which has transcription enabled, performs a consultation with Agent 2 from Queue B, then Agent 2 will be considered as a transcription user because Agent2 was involved in the conversation even if the consult transfer was not completed.
• When a call that is set for voice transcription is transferred to a communicate user, transcription continues and the communicate user is billed for transcription by the minute.
• When an ACD agent consult transfer is performed and the customer is listening to music (without vocals), the music may be transcribed as filler words (for example, um, ah, and so on) and a Conference icon appears in the transcript even though the music is not a conference between multiple agents.
• Consult segments are not processed as part of acoustic analysis. They are not included in the acoustic analysis percentage calculation.
• When the interaction is put on hold and the customer is listening to music (without vocals), the music may be transcribed as filler words (for example, um, ah, and so on) and an Agent icon appears in the transcript even though the agent is not talking.
• When an interaction is transferred to a queue,  the in-queue flow may not be fully executed prior to delivering the interaction to an available agent. As a result, if the in-queue flow attempts to change the language, the language change or switch may not take effect since the transcription flow may not have been executed prior to the transfer completing.
• BYOC Premises post call voice transcription does not transcribe calls beyond a 10-hour limit.
• BYOC Premises post call voice transcription requires dual channel recording.
• BYOC Premises post call voice transcription of recordings is only supported with audio in the following formats: Opus, PCMU, PCMA or L16.
• The list of dialects/languages in external trunk configuration is limited, as a results, for some dialects to be transcribed, the dialect must first be set through an Architect flow. For more information, see External trunk settings.
• When transferring calls to external parties, there may be instances where the participant associated with the transcript phrase might not be accurately reflected.
• Sentiment Feedback limits phrases to 350 characters for Korean, Chinese, and Japanese dialects.

For more information, see About voice transcription.