Music, Voice, SFX, Soundtracks: Pika expands into audio

Pika brings together four audio models dedicated to music, sound effects, voice, and soundtracks synchronized with video.

Pika is expanding its generative media environment into audio with a family of four specialized models. Pika Soundtrack creates a soundtrack from video, Pika Music generates complete songs, Pika SFX focuses on sound effects, and Pika Speech handles expressive text-to-speech and voice cloning. For now, the full family is available exclusively through the Pika API Club.

Pika Soundtrack analyzes video directly to generate music, ambience, voice, and sound effects synchronized with the action. The prompt can be left blank to produce a complete soundscape or used to specify which elements should be emphasized, included, or excluded. In its own tests against LTX-2.3 Foley V2A, HunyuanVideo-Foley, and MMAudio v2, Pika reports the strongest semantic alignment and lowest audiovisual desynchronization, with 0.617 seconds of compute per generated second of video.

Pika Music accepts several types of inputs that can be combined, including text prompts, lyrics, vocal references, and reference tracks. It can generate songs up to six minutes long, with the ability to combine these different signals within the same generation. Pika says a 90-second song takes an average of 6.21 seconds to generate in its tests, or roughly 14.5 times faster than real-time playback.

Pika SFX turns a description into a sound effect up to 20 seconds long in 44.1 kHz stereo. The model can handle either a single event or a longer sequence while accounting for material, space, perspective, timing, or the requested texture. The company reports an average end-to-end generation time of 0.847 seconds in its local benchmark.

Pika Speech completes the lineup with 48 kHz text-to-speech. It can use preset voices or create a voice clone from a few seconds of reference audio, while a description can guide pacing, intonation, or timbre. Requests can be up to five minutes long, and Pika reports a real-time factor of 0.02 in its local tests, equivalent to roughly one second of compute for one minute of speech.

Pricing is the other major point highlighted by Pika. The company says its audio models can cost up to 20 times less than some alternatives, depending on the category. It compares Pika Speech with ElevenLabs v3, Cartesia, ElevenLabs Turbo, and Fish Audio, while Pika Music is compared with several competing music models. These differences are based on list prices collected by Pika and should therefore be read as company-published comparisons rather than an independent pricing benchmark.