Files
muninn/app/Support/VideoWorkflow/StepHandlers/TtsVoiceoverHandler.php
T
m105andClaude Sonnet 5 55fa2ebfda Add script/TTS step handlers, dashboard, and pipeline UI (#35, #36, #43, #44)
StepHandler abstraction (app/Support/VideoWorkflow/StepHandlers/) plugs
into WorkflowEngine without coupling it to how any given step type is
actually executed:
- ScriptGenerationHandler: builds a prompt from the project's title/
  description (or an explicit outline/tone/keywords/target length in
  step input), generates via an active OpenAI AiProvider, stores the
  result as a 'script' asset.
- TtsVoiceoverHandler: voices step.input.text, or falls back to the
  project's latest script asset, via an active ElevenLabs AiProvider;
  voice_id/model_id are configurable per step. No direct "speed"
  control — ElevenLabs' base TTS endpoint doesn't expose one.

RunWorkflowStepJob resolves a step's handler by type and runs it on
the queue, matching the existing PublishPostJob pattern (including a
stale-dispatch guard: no-op if the step isn't 'in_progress' anymore).

UI, under its own /video-projects routes and nav entry — separate
from /projects per the earlier decision:
- VideoProjects/Index (#43): list projects with status and step
  progress, create new ones.
- VideoProjects/Show (#44): start a project from a WorkflowTemplate or
  a hardcoded default 8-step pipeline; per-step Run/Approve/Reject/
  Retry/Skip actions; auto-queues the next step after any transition
  when a handler exists for its type (steps with no handler yet just
  sit in_progress rather than auto-failing); lists project assets.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-08-06 20:56:01 +03:00

96 lines
2.7 KiB
PHP

<?php
namespace App\Support\VideoWorkflow\StepHandlers;
use App\Models\AiProvider;
use App\Models\WorkflowStep;
use App\Support\AiProviders\AiGenerationException;
use App\Support\AiProviders\AiProviderManager;
use App\Support\AssetStorage\AssetStorage;
use App\Support\VideoWorkflow\WorkflowEngine;
/**
* Synthesizes narration audio via an ElevenLabs-backed AiProvider. Text
* comes from $step->input['text'] if given, otherwise the project's latest
* 'script' asset. Voice selection (input.voice_id / input.model_id) is
* per-step, falling back to the AiProvider's own config.
*
* Note: ElevenLabs' base TTS endpoint doesn't expose a direct "speed"
* parameter — only voice_id/model_id are wired through here for now.
*/
class TtsVoiceoverHandler implements StepHandler
{
public function __construct(
private readonly AiProviderManager $aiProviders,
private readonly AssetStorage $assets,
) {}
public function type(): string
{
return 'tts_voiceover';
}
public function handle(WorkflowStep $step, WorkflowEngine $engine): void
{
$provider = AiProvider::query()->where('provider', 'elevenlabs')->where('is_active', true)->first();
if (! $provider) {
$engine->failStep($step, 'No active ElevenLabs provider configured.');
return;
}
$text = $this->resolveText($step);
if (! $text) {
$engine->failStep($step, 'No script text available to voice — set input.text or complete a script_generation step first.');
return;
}
$input = $step->input ?? [];
$options = array_filter([
'voice_id' => $input['voice_id'] ?? null,
'model_id' => $input['model_id'] ?? null,
]);
try {
$result = $this->aiProviders->generate($provider, $text, $options);
} catch (AiGenerationException $e) {
$engine->failStep($step, $e->getMessage());
return;
}
$asset = $this->assets->store(
$step->videoProject,
'audio',
$result->content,
'mp3',
step: $step,
mimeType: $result->contentType,
);
$engine->completeStep($step, [
'asset_id' => $asset->id,
'characters' => strlen($text),
]);
}
private function resolveText(WorkflowStep $step): ?string
{
$input = $step->input ?? [];
if (! empty($input['text'])) {
return $input['text'];
}
$scriptAsset = $step->videoProject->assets()
->where('type', 'script')
->orderByDesc('version')
->first();
return $scriptAsset ? $this->assets->contents($scriptAsset) : null;
}
}