Speech and text driven HMM-based body animation synthesis

• US 8,224,652 B2
• Filed: 09/26/2008
• Issued: 07/17/2012
• Est. Priority Date: 09/26/2008
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Wang, Lijuan, Ma, Lei, Soong, Frank Kao-Ping
• Primary Examiner(s): Azad, Abul
• Application Number:
  US12/239,564
• Publication Number:
  US 20100082345A1
• Time in Patent Office: 1,390 Days
• Field of Search: 704/235, 704/260, 704/257, 704/270, 704/275
• US Class Current: 704/275
• CPC Class Codes: G06T 13/205 driven by audio data
• CPC Class Codes: G10L 13/00 Speech synthesis; Text to speech systems
• CPC Class Codes: G10L 21/06 Transformation of speech into a non-audible representation, e.g. speech visualisation or speech processing for tactile aids G10L15/26 takes precedence

Abstract

An “Animation Synthesizer” uses trainable probabilistic models, such as Hidden Markov Models (HMM), Artificial Neural Networks (ANN), etc., to provide speech and text driven body animation synthesis. Probabilistic models are trained using synchronized motion and speech inputs (e.g., live or recorded audio/video feeds) at various speech levels, such as sentences, phrases, words, phonemes, sub-phonemes, etc., depending upon the available data, and the motion type or body part being modeled. The Animation Synthesizer then uses the trainable probabilistic model for selecting animation trajectories for one or more different body parts (e.g., face, head, hands, arms, etc.) based on an arbitrary text and/or speech input. These animation trajectories are then used to synthesize a sequence of animations for digital avatars, cartoon characters, computer generated anthropomorphic persons or creatures, actual motions for physical robots, etc., that are synchronized with a speech output corresponding to the text and/or speech input.

Claims

1. A method for synthesizing animation motions, comprising steps for:

providing a set of one or more animation models, each animation model providing a set of probabilistic motions of one or more body parts learned from acoustic features and speech prosody information extracted from a set of one or more audio/video training signals comprising synchronized speech and body motions;

wherein the probabilistic motions of the animation models are constructed as individual “

animation units”

by modeling Euler angles of each modeled body part as a function of corresponding acoustic features and speech prosody information and generating full context labels for each animation unit;

wherein two or more of the animation models are clustered using predictive analysis techniques to allow selection of animation units from multiple animation models to ensure that a smooth animation trajectory can be constructed from any selected animation units;

receiving an arbitrary speech input synthesized from an arbitrary text input including one or more “

emotion tags”

associated with one or more portions of the arbitrary text input, said emotion tags being used to change emotional characteristics of the synthesized arbitrary speech input;

evaluating the speech input to extract a set of acoustic features and speech prosody information from the speech input;

predicting a sequence of the probabilistic motions to construct an animation trajectory which best explains the arbitrary speech input, by applying one or more of the set of animation models to the set of acoustic features and speech prosody information extracted from the arbitrary speech input; and

generating an animation sequence from the predicted sequence of the probabilistic motions.

2. The method of claim 1 wherein the animation sequence is used to construct an audio/video animation of an avatar, said animation including the arbitrary speech input synchronized to the animation sequence.

3. The method of claim 1 wherein the animation sequence is used to construct a robot control sequence, said robot control sequence including the arbitrary speech input synchronized to robotic motions corresponding to the animation sequence.

4. The method of claim 1 wherein the acoustic features and speech prosody information includes one or more of speech F0 measurements, speech energy measurements, speech timing measurements, part-of-speech, and speech semantics.

5. The method of claim 1 wherein each of the probabilistic motions correspond to one or more separate speech levels, including sentences, phrases, words, phonemes, and sub-phonemes.

6. The method of claim 1 wherein each of the probabilistic motions include one or more of a mouth, nose, eyes, eyebrows, ears, face, head, fingers, hands, arms, legs, feet, torso, spine, and skeletal elements of a body.

7. The method of claim 1 further comprising translating the arbitrary text input from a first language to a second language prior to synthesizing the arbitrary speech input from the translated arbitrary text input.

8. The method of claim 1 wherein the arbitrary speech input is synthesized from a real speech input of a user.

9. A system for generating audio/video animations from an arbitrary speech input, comprising:

a device for training one or more probabilistic motion models from one or more audio/video training signals comprising synchronized speech and body part motions;

wherein training each probabilistic motion model further comprises learning a set of animation units corresponding to actual motions of specific body parts relative to acoustic features and speech prosody information extracted from one or more of the audio/video training signals;

wherein the animation units are constructed by modeling Euler angles of each modeled body part as a function of corresponding acoustic features and speech prosody information and generating full context labels for each animation unit;

wherein two or more of the probabilistic motion models are clustered using predictive analysis techniques to allow selection of animation units from multiple probabilistic motion models to ensure that a smooth animation trajectory can be constructed from any selected animation units;

a device for receiving an arbitrary text input having one or more “

emotion tags”

associated with one or more portions of the arbitrary text input;

a device for synthesizing an arbitrary speech input from the from the arbitrary text input, and wherein the emotion tags are used to change emotional characteristics of the associated portions of the synthesized arbitrary speech input;

a device for extracting a set of acoustic features and speech prosody information from the arbitrary speech input;

a device for predicting a sequence of the animation units to construct an animation trajectory which probabilistically explains the arbitrary speech input, by applying one or more of the set of motion models to the set of acoustic features and speech prosody information extracted from the arbitrary speech input;

a device for generating an animation sequence from the predicted sequence of the animation units; and

a device for constructing an audio/video animation of an avatar, said animation including the arbitrary speech input synchronized to the animation sequence.

10. The system of claim 9 wherein each of the animation units correspond to one or more separate speech levels, including sentences, phrases, words, phonemes, and sub-phonemes.

11. The system of claim 9 further comprising a device for translating the arbitrary text input from a first language to a second language prior to synthesizing the arbitrary speech input from the translated arbitrary text input.

12. The system of claim 9 further comprising one or more multi-point video conferencing devices, wherein each of two or more users is represented to each of one or more other users by a real-time audio/video animation of an avatar generated from real-time arbitrary speech inputs of each corresponding user.

13. A computer-readable medium, not comprising modulated data signals and carrier waves, having computer executable instructions stored therein for constructing an audio/video animation of an avatar as a function of an arbitrary text input, said instructions comprising:

learning one or more probabilistic motion models from one or more audio/video training signals, wherein each audio video training signal includes synchronized speech and body part motions of a human speaker;

wherein learning each probabilistic motion model further comprises learning a set of animation units corresponding to actual motions of specific body parts of the human speaker as a predictive function of acoustic features and speech prosody information extracted from one or more of the audio/video training signals;

wherein the animation units are constructed by modeling Euler angles of each modeled body part as a function of corresponding acoustic features and speech prosody information and generating full context labels for each animation unit;

wherein two or more of the probabilistic motion models are clustered using predictive analysis techniques to allow selection of animation units from multiple probabilistic motion models to ensure that a smooth animation trajectory can be constructed from any selected animation units;

receiving an arbitrary text input including one or more “

emotion tags”

associated with one or more portions of the arbitrary text input;

synthesizing a speech signal from the arbitrary text input, and wherein the emotion tags are used to change emotional characteristics of the associated portions of the synthesized speech signal;

extracting a set of acoustic features and speech prosody information from the synthesized speech signal;

predicting a sequence of the animation units to construct an animation trajectory which probabilistically explains the synthesized speech signal, by applying one or more of the set of motion models to the set of acoustic features and speech prosody information extracted from the synthesized speech signal; and

constructing an audio/video animation of an avatar from the predicted sequence of the animation units;

said animation including the synthesized speech signal synchronized to the animation.

14. The computer-readable medium of claim 13 wherein each of the animation units correspond to one or more separate speech levels, including sentences, phrases, words, phonemes, and sub-phonemes.

15. The computer-readable medium of claim 13 wherein the arbitrary text input is translated from a first language to a second language prior to synthesizing the speech signal.

16. The computer-readable medium of claim 13 wherein the arbitrary text input is extracted from an email message, and wherein the audio/video animation of the avatar represents a digital avatar speaking the text of the email message while exhibiting body part motions corresponding to animation units modeled after one or more real human speakers.