Cobalt VoiceBio vs Recognition Technologies Speaker Recognition

Cobalt VoiceBio provides voice biometrics for speaker verification and identification, including text-independent operation. It converts speech into voiceprints representing a speaker independently of the words spoken. StreamingVerify compares live audio with one voiceprint and returns a similarity score and match decision; StreamingIdentify compares live audio with a list and returns the best match or none. StreamingEnroll creates voiceprints and reports how much more audio is needed to complete enrollment. Voiceprints can be compared without resending audio or vectorized for indexing and downstream tasks. The engine runs on customer hardware, keeping audio and voiceprints inside that infrastructure. Cobalt lists embedded, on-premise, and private cloud deployment options. VoiceBio serves a gRPC API and includes pre-generated Go and Python SDKs, with instructions for generating SDKs for other languages. The trial supports 8 kHz and 16 kHz speech and is listed at free for 100 days; Arm64 is not included. Listed system requirements are an x86_64 CPU, 500 MB disk space, and 2 GB RAM. Cobalt describes using VoiceBio to authenticate customers, employees, and partners, and says it can be combined with passcodes or other credentials. Pricing is available on request.

Recognition Technologies Speaker Recognition is a language- and text-independent voice biometrics system for identifying and verifying speakers. A person may enroll in one language and be identified or verified in another. The engine also supports classification, detection, tracking and segmentation. Identification can use open-set or closed-set approaches; closed-set identification can return the nearest enrolled voice even if the speaker is not in the database. The maker describes verification enrollment as about 30 seconds of speech, followed by a few seconds when the speaker returns. Interfaces include a C++ SDK/API, command line, web service, Android and iOS; the engine can also be used as a module of the maker’s IVR system. Deployment options include embedded, server-based, standalone, hosting at Recognition Technologies or hosting at the customer’s facilities. The product page lists Linux, Apple Macintosh OS X and Windows support. Its telephony version supports speaker verification and identification for telephone authentication and access-level control. Pricing is on request, and an evaluation account for the hosted version may be available to interested organizations.