Volcano Engine has directly upgraded and renamed its previous-generation "Doubao Speech Synthesis Model 2.0" into the "Doubao Audio Generation Model 1.0 (Seed-Audio 1.0)." Going from "speech synthesis" to "audio generation" is not a renaming game — it means one prompt can spit out character dialogue, ambient sound effects, and background music as a whole package, instead of generating character A first, then character B, then layering in the BGM, and finally dragging everything into an editor and aligning it track by track. This piece uses three hands-on cases to show what it can do and where the limits are.

What Seed-Audio 1.0 Is

Think of it as "the AI version of the entire voice-over post-production workflow." The traditional pipeline: hire a voice actor to record lines, hire a sound designer to lay in ambience, hire a composer for the BGM, and finally have a mixing engineer blend it all. Seed-Audio 1.0 compresses that line into a single prompt — you describe a scene, and it packages and outputs the voices, effects, and score directly. The core upgrade's official name: "film-grade all-element direct output."

Core Features

Feature One: Long-Form Continuation, with Voice Consistent Across Segments

The "designer monologue" segment from the previous 2.0 version was kept intact, and 1.0 was asked to continue it — the whole clip runs 1 minute 10 seconds. The first 16 seconds are the original monologue; from second 16 it moves into his dialogue with the client — same character, same voice, same exhausted state, but the scene shifts from a solo monologue to a two-person face-off.

The key detail: the "beep... beep... beep..." busy tone after the phone hangs up and the three seconds of dead silence were generated by the AI itself, with no post-production layering at all. That is what "all-element direct output" is worth — it understands what kind of sound rhythm a narrative needs.

Feature Two: One Prompt Generates a Complete Animated-Short Soundtrack

Test scenario: a script for a three-character animated short — narrator (young male), elder (elderly male), and a boy. The lines carry heavy emotional tension: the narrator has a deep, mellow Chinese-style animation tone; the elder's voice is aged and raspy with a condescending contempt; the boy's voice is clear and bright with anger in it.

Beyond the voices, the script also calls for guzheng, drums, strings, shuffling footsteps, a spirit sword unsheathing, metal strikes, crowd laughter, and bell chimes — everything a power-fantasy short should have.

The old workflow needed: generate each character separately → find BGM → layer footsteps, palm-wind sounds, brazier sounds → drag into an editor and align track by track. Seed-Audio 1.0 spits out the whole soundscape the animated short should have from a single prompt.

Feature Three: Live Sports Commentary, Also Direct Output

Using the real World Cup backdrop of "Cape Verde's goalkeeper shut out Spain," we had it generate a commentary segment. Sports broadcasting doesn't call for pre-arranged cinematic sound design but for chaotic in-the-moment texture: the crowd roaring, stadium echo, the commentator riding the rhythm of the match — holding back, speeding up, bursting out, falling back.

Listening to the result, the layers are distinct: the voice up front, the stadium sounds behind, the background crowd never drowning out the commentary — it sounds like sitting in the broadcast booth.

Tested with an animated-short script for the multi-character scene: each character's voice, emotion, and spatial position were all reproduced in one pass.

Hands-On Experience

Strengths

• Saves post-production: the most immediate impression is no more opening a DAW (digital audio workstation) to align tracks. One prompt in, complete sound out.
• Emotion has drama: it doesn't just sound human; it can "direct" a scene with sound. The sleepy irritation of a client boss woken by a call, the busy tone after the phone hangs up — both are part of the narrative.
• Convincing live texture: for scenes like sports commentary that need chaotic layering (voice + crowd + echo), the layers stay distinct instead of smearing together.

Limitations

• Complex long passages degrade toward the end: like video generation, the longer and more complex a passage gets, the more fidelity drops; you'll need to generate in segments or fine-tune manually.
• Demands good prompt craft: "all-element direct output" raises the bar on descriptive ability — you must spell out the characters, scene, emotional arc, and which effects you need, or the model will improvise.

Use Cases

• Audiobooks / animated shorts / micro-drama dubbing: the community had already been asking "when does this plug into Tomato Novel"; all-element direct output makes industrial-scale audio content possible
• Short-video sound post: vlogs and commentary videos get ambience and BGM produced together
• Sports / news live commentary: content that needs live texture and emotional layering
• Game cutscene dubbing: NPC dialogue, combat effects, and background music in one shot

The API is in invitation-only beta; apply through the Volcano Ark console.