ControlAudio: Tackling Text-Guided, Timing-Indicated
and Intelligible Audio Generation via Progressive Diffusion Modeling

Abstract

Text-to-audio (TTA) generation with fine-grained control signals, e.g., precise timing control or intelligible speech content, has been explored in recent works. However, constrained by data scarcity, their generation performance at scale is still compromised. In this study, we recast controllable TTA generation as a multi-task learning problem and introduce a progressive diffusion modeling approach, ControlAudio. Our method adeptly fits distributions conditioned on more fine-grained information, including text, timing, and phoneme features, through a step-by-step strategy. First, we propose a data construction method spanning both annotation and simulation, augmenting condition information in the sequence of text, timing, and phoneme. Second, at the model training stage, we pretrain a diffusion transformer (DiT) on large-scale text-audio pairs, achieving scalable TTA generation, and then incrementally integrate the timing and phoneme features with unified semantic representations, expanding controllability. Finally, at the inference stage, we propose progressively guided generation, which sequentially emphasizes more fine-grained information, aligning inherently with the coarse-to-fine sampling nature of DiT. Extensive experiments show that ControlAudio achieves state-of-the-art performance in terms of temporal accuracy and speech clarity, significantly outperforming existing methods on both objective and subjective evaluations.

Timing-Controlled Audio Generation

Text prompt A bathtub is being filled or washed. Timing prompt Bathtub (filling or washing): 0.0s - 6.598s Bathtub (filling or washing): 6.748s - 10.0s
Structured prompt A bathtub is being filled or washed. @{Bathtub (filling or washing) <0.00,6.60><6.75,10.00>}
ControlAudio	AudioLDM	Tango	Stable Audio*	AudioComposer	Ground Truth

Text prompt A woman sings with music and an alarm clock sounds. Timing prompt A woman sings: 0.00s - 6.73s Alarm clock: 6.73s - 10.00s
Structured prompt A woman sings with music and an alarm clock sounds. @{A woman sings & <0.00,6.73>}@{Alarm clock & <6.73,10.00>}
ControlAudio	AudioLDM	Tango	Stable Audio*	AudioComposer	Ground Truth

Text prompt A telephone rings repeatedly. Timing prompt Telephone bell ringing: 0.0s - 2.221s Telephone bell ringing: 2.568s - 8.022s Telephone bell ringing: 8.473s - 10.0s
Structured prompt A telephone rings repeatedly. @{Telephone bell ringing & <0.00,2.22><2.57,8.02><8.47,10.00>}
ControlAudio	AudioLDM	Tango	Stable Audio*	AudioComposer	Ground Truth

Text prompt A man is speaking and an engine is running. Timing prompt Engine is running : 1.40s - 10.00s Male speech, man speaking : 0.00s - 1.40s LLM Planning: "The engine’s running, let’s go."
Structured prompt A man is speaking and an engine is running. @{Engine is running <0.00,10.00>}@{Male speech, man speaking <0.00,1.40><DH><AH0>< ><EH1><N><JH><AH0><N><Z>< ><R><AH1><N><IH0><NG>< ,>< ><L><EH1><T><S>< ><G><OW1>< .>}
ControlAudio	AudioLDM	Tango	Stable Audio*	AudioComposer	Ground Truth

Text prompt Music plays, followed by mechanisms, typing, beeps, and an alarm. Timing prompt Music : 0.00s - 10.00s Beeps : 1.00s - 1.20s 3.00s - 3.20s 4.90s - 5.10s 6.90s - 7.10s Typing : 1.20s - 7.80s Alarm : 7.85s - 8.50s
Structured prompt Music plays, followed by mechanisms, typing, beeps, and an alarm. @{Music. & <0.00,10.00>}@{Beeps. & <1.00,1.20><3.00,3.20><4.90,5.10><6.90,7.10>}@{Typing. & <1.20,7.80>}@{Alarm. & <7.85,8.50>}
ControlAudio	AudioLDM	Tango	Stable Audio*	AudioComposer	Ground Truth

Intelligible Audio Generation

Text prompt People talking with the dull roar of a vehicle on the road. Content prompt What is she doing? What the fuck? What the fuck? She's in the middle of the road. Mate, he's on a joyride.
ControlAudio	AudioLDM 2 Speech	VoiceLDM-S	VoiceLDM-M	Ground Truth

Text prompt A gun cocking followed by a gunshot and a man talking. Content prompt There you see? No problem.
ControlAudio	AudioLDM 2 Speech	VoiceLDM-S	VoiceLDM-M	Ground Truth

Text prompt A man speaking over an intercom as a crowd of people talk followed by a dog barking. Content prompt and contain them until that person can be taken into custody effectively and safely on the part of the other team of police sheriffs.
ControlAudio	AudioLDM 2 Speech	VoiceLDM-S	VoiceLDM-M	Ground Truth

Text prompt Females voice narrating a scene as music is playing and rain drops are falling. Content prompt Daniel came out of the airport. He raised one arm to hail a taxi.
ControlAudio	AudioLDM 2 Speech	VoiceLDM-S	VoiceLDM-M	Ground Truth

Text prompt Splashing water followed by a girl speaking then scraping and spitting. Content prompt This is the last time you did that first thing. Same thing.
ControlAudio	AudioLDM 2 Speech	VoiceLDM-S	VoiceLDM-M	Ground Truth