Bloomberg Vault adds BSpeech for multi-language voice transcription

Bloomberg has added its BSpeech voice transcription service to Bloomberg Vault, enabling automatic conversion of recorded voice calls into searchable transcripts in more than 50 languages. The capability is available as a premium add-on for Vault clients that archive voice communications.

Transcripts are generated during the archiving process and appear in Vault’s search and surveillance environment so voice recordings can be searched, supervised, analyzed and audited alongside chat and email. Bloomberg said the feature allows firms to integrate voice data into existing compliance workflows without separate systems.

BSpeech uses machine learning models trained on Bloomberg’s financial data corpus to improve recognition of industry terms, stock tickers and proper names. Bloomberg said the models will be refined periodically to better reflect the language used in financial conversations.

Perry Goetz, Global Head of Compliance Solutions (Vault) at Bloomberg, said: “As regulatory expectations around voice surveillance continue to rise, firms are under increasing pressure to apply the same level of oversight to voice as they do to written channels. By enabling compliance teams to monitor voice communications with the same rigor as written channels, surveillance efficiency is enhanced, manual review burdens are reduced and firms’ ability to detect risk across communications improves.”

The integration runs on Bloomberg’s secure, scalable infrastructure and is offered as an enterprise option that can be activated for clients using Vault’s Corporate Voice Capture & Archive services. For firms already capturing voice for record-keeping, transcription can be turned on within existing workflows rather than requiring separate systems.

Bloomberg said searchable transcripts will let compliance teams run automated searches, apply surveillance rules and include voice data in analytics that cover multiple communication channels. Archived audio files will be preserved alongside the generated text.