speaker verification
In plain wordsChecking whether a voice really belongs to the person it claims to be, as when a bank confirms your identity by the sound of your voice on the phone.
The biometric task of deciding whether a voice sample belongs to the enrolled person whose identity is claimed. At enrollment a model computes a speaker embeddingembedding, n.: A list of numbers that stands for a word, picture or other item, arranged so that similar items get similar lists. It lets a computer measure similarity of meaning as distance. from sample utterances; at test time it embeds the new sample, scores the pair with cosine similaritycosine similarity, n.: A score from −1 to 1 for how closely two lists of numbers point the same way, ignoring their size. It is the standard test of whether two texts mean similar things. or probabilistic linear discriminant analysis, and accepts the claim if the score clears a threshold. As a one-to-one check it differs from speaker identification, a one-to-many search among known voices, and the two together form speaker recognition; either can be text-dependent (a fixed or prompted passphrase) or text-independent. Speaker representations evolved from Gaussian-mixture models adapted from a universal background model (2000) and i-vectors (2011) to neural d-vectors and x-vectors (2014–2018) and ECAPA-TDNN (2020). Performance is reported as equal error rate (where false acceptances equal false rejections) and as detection cost, on benchmarks such as the NIST Speaker Recognition Evaluations and VoxCeleb. Replayed recordings and synthetic speech from voice cloningvoice cloning, n.: Using AI to copy a specific person's voice from recordings, so a computer can say new things that sound like them — useful for dubbing and accessibility, and abused in scams. threaten voice authentication, so deployed systems add liveness and spoofing detection, studied in the ASVspoof challenges since 2015.
After the cloned-voice incident, the bank added spoofing detection on top of speaker verification.
Origin
Lawrence Kersta of Bell Labs promoted visual 'voiceprint' comparison of spectrograms in 1962; automatic speaker recognition and verification were studied at Bell Labs, Texas Instruments and elsewhere from the 1960s. The US NIST has run Speaker Recognition Evaluations since 1996, and neural embeddings dominated from the late 2010s.
Usage note
Voice templates used to identify people count as biometric data under laws such as the EU's GDPR and Illinois's Biometric Information Privacy Act. Not to be confused with speech recognitionspeech recognition, n.: Technology that turns spoken words into written text, as when a phone transcribes a voice message or captions appear automatically under a video., which concerns what was said rather than who said it.
Related
- speaker diarizationspeaker diarization, n.: Working out who spoke when in a recording, splitting a meeting or phone call into turns labeled Speaker 1, Speaker 2 and so on, without necessarily knowing who the speakers are.
- voice cloningvoice cloning, n.: Using AI to copy a specific person's voice from recordings, so a computer can say new things that sound like them — useful for dubbing and accessibility, and abused in scams.
- deepfake detectiondeepfake detection, n.: Using software, often itself AI, to tell whether a video, photo or voice recording has been faked or generated by AI rather than captured from reality.
- embeddingembedding, n.: A list of numbers that stands for a word, picture or other item, arranged so that similar items get similar lists. It lets a computer measure similarity of meaning as distance.
- facial recognitionfacial recognition, n.: Technology that identifies or checks who someone is by comparing a picture of their face with stored photos, used by phones, passport gates and police databases.
- MFCCMFCC, abbr.: A short list of numbers, usually about 13, that summarizes the shape of a brief slice of sound in a way that mimics human hearing; for decades the standard input to speech recognizers.
- speech recognitionspeech recognition, n.: Technology that turns spoken words into written text, as when a phone transcribes a voice message or captions appear automatically under a video.
Sources
- Snyder, D., Garcia-Romero, D., Sell, G., Povey, D. & Khudanpur, S. (2018). X-Vectors: Robust DNN Embeddings for Speaker Recognition. ICASSP 2018.
- Nagrani, A., Chung, J. S. & Zisserman, A. (2017). VoxCeleb: A Large-Scale Speaker Identification Dataset. Interspeech 2017.
Mentioned in
speech recognitionspeech recognition, n.: Technology that turns spoken words into written text, as when a phone transcribes a voice message or captions appear automatically under a video., voice cloningvoice cloning, n.: Using AI to copy a specific person's voice from recordings, so a computer can say new things that sound like them — useful for dubbing and accessibility, and abused in scams., biometric databiometric data, n.: Information about your body or behavior — your face, fingerprints, voice, eyes or the way you walk — measured by technology and used to recognize who you are or to guess things about you., deepfake detectiondeepfake detection, n.: Using software, often itself AI, to tell whether a video, photo or voice recording has been faked or generated by AI rather than captured from reality., speaker diarizationspeaker diarization, n.: Working out who spoke when in a recording, splitting a meeting or phone call into turns labeled Speaker 1, Speaker 2 and so on, without necessarily knowing who the speakers are., wake wordwake word, n.: A word or phrase, such as 'Alexa' or 'Hey Google', that a smart speaker or phone listens for constantly and that switches it into full listening mode., metric learningmetric learning, n.: Teaching a model to measure similarity: it learns to place items so that ones that should count as alike end up close together and different ones far apart., MFCCMFCC, abbr.: A short list of numbers, usually about 13, that summarizes the shape of a brief slice of sound in a way that mimics human hearing; for decades the standard input to speech recognizers., time delay neural networktime delay neural network, n.: An early network for speech that applied the same small pattern detector at every moment of a sound, so it could recognize a sound wherever it occurred in time.