dobetternorge-tools/includes/GcpSpeechClient.php

<?php
/**
 * Google Cloud Speech-to-Text v2 — no SDK, raw curl + API key auth.
 *
 * Credential priority:
 *   1. GCP_PROJECT_ID / GCP_API_KEY env vars
 *   2. /etc/bnl/gcp.php  (array with 'project_id' and 'api_key')
 */
class GcpSpeechClient {
    private string $projectId;
    private string $apiKey;

    private const LOCALE_MAP = [
        'no' => 'nb-NO', 'nb' => 'nb-NO', 'nn' => 'nn-NO',
        'en' => 'en-US', 'sv' => 'sv-SE', 'da' => 'da-DK',
        'de' => 'de-DE', 'fr' => 'fr-FR',
    ];

    public function __construct(string $projectId, string $apiKey) {
        $this->projectId = $projectId;
        $this->apiKey    = $apiKey;
    }

    public static function fromConfig(): ?self {
        $projectId = getenv('GCP_PROJECT_ID') ?: null;
        $apiKey    = getenv('GCP_API_KEY')    ?: null;

        if (!$projectId || !$apiKey) {
            $cfg = @include '/etc/bnl/gcp.php';
            if (is_array($cfg)) {
                if (!$projectId && !empty($cfg['project_id'])) $projectId = (string)$cfg['project_id'];
                if (!$apiKey    && !empty($cfg['api_key']))    $apiKey    = (string)$cfg['api_key'];
            }
        }

        if (!$projectId || !$apiKey) return null;
        return new self($projectId, $apiKey);
    }

    /**
     * Transcribe an audio file using GCP Speech-to-Text v2.
     *
     * Returns a Whisper-compatible array on success:
     *   ['text', 'language', 'duration', 'segments', 'num_speakers']
     * Returns null on any failure (caller should fall back to Whisper).
     */
    public function transcribe(
        string $audioPath,
        string $mimeType,
        string $language,
        bool   $diarize,
        int    $minSpeakers = 2,
        int    $maxSpeakers = 6,
        int    $timeoutSec  = 270
    ): ?array {
        $locale = $this->resolveLocale($language);

        $features = ['enableAutomaticPunctuation' => true];
        if ($diarize) {
            $features['diarizationConfig'] = [
                'minSpeakerCount' => max(2, $minSpeakers),
                'maxSpeakerCount' => max(max(2, $minSpeakers), $maxSpeakers),
            ];
        }

        $langCodes = [$locale];
        // Add Nynorsk as secondary when processing Norwegian content
        if ($locale === 'nb-NO') $langCodes[] = 'nn-NO';

        $config = [
            'autoDecodingConfig' => (object)[],
            'languageCodes'      => $langCodes,
            'model'              => 'long',
            'features'           => $features,
        ];

        $audioContent = base64_encode((string)file_get_contents($audioPath));
        if ($audioContent === '') {
            error_log("GcpSpeechClient: failed to read audio file");
            return null;
        }

        $body = json_encode(['config' => $config, 'content' => $audioContent]);

        $url = sprintf(
            'https://speech.googleapis.com/v2/projects/%s/locations/global/recognizers/_:recognize?key=%s',
            rawurlencode($this->projectId),
            rawurlencode($this->apiKey)
        );

        $ch = curl_init($url);
        curl_setopt_array($ch, [
            CURLOPT_POST           => true,
            CURLOPT_POSTFIELDS     => $body,
            CURLOPT_HTTPHEADER     => ['Content-Type: application/json'],
            CURLOPT_RETURNTRANSFER => true,
            CURLOPT_TIMEOUT        => 60,
        ]);
        $response = curl_exec($ch);
        $httpCode = curl_getinfo($ch, CURLINFO_HTTP_CODE);
        $curlErr  = curl_error($ch);
        curl_close($ch);

        if ($curlErr || !is_string($response)) {
            error_log("GcpSpeechClient: curl error: {$curlErr}");
            return null;
        }
        if ($httpCode !== 200) {
            error_log("GcpSpeechClient: HTTP {$httpCode}: " . substr($response, 0, 300));
            return null;
        }

        $data = json_decode($response, true);
        if (!is_array($data)) return null;

        // Long audio: GCP returns an operation name to poll
        if (isset($data['name']) && str_contains((string)$data['name'], '/operations/')) {
            $data = $this->pollOperation((string)$data['name'], $timeoutSec);
            if ($data === null) return null;
        }

        if (empty($data['results']) || !is_array($data['results'])) {
            error_log("GcpSpeechClient: no results in response");
            return null;
        }

        return $this->normalizeResults($data['results'], $locale);
    }

    private function pollOperation(string $operationName, int $timeoutSec): ?array {
        $url      = "https://speech.googleapis.com/v2/{$operationName}?key=" . rawurlencode($this->apiKey);
        $deadline = time() + $timeoutSec;
        $interval = 5;

        while (time() < $deadline) {
            sleep($interval);
            $interval = min($interval + 5, 15);

            $ch = curl_init($url);
            curl_setopt_array($ch, [
                CURLOPT_RETURNTRANSFER => true,
                CURLOPT_TIMEOUT        => 15,
            ]);
            $response = curl_exec($ch);
            $httpCode = curl_getinfo($ch, CURLINFO_HTTP_CODE);
            curl_close($ch);

            if ($httpCode !== 200 || !is_string($response)) continue;

            $data = json_decode($response, true);
            if (!is_array($data)) continue;
            if (!empty($data['done'])) {
                return $data['response'] ?? null;
            }
        }

        error_log("GcpSpeechClient: operation timed out after {$timeoutSec}s");
        return null;
    }

    private function normalizeResults(array $gcpResults, string $locale): ?array {
        $transcriptParts = [];
        $allWords        = [];

        foreach ($gcpResults as $result) {
            $alt = $result['alternatives'][0] ?? null;
            if (!$alt) continue;
            if (!empty($alt['transcript'])) $transcriptParts[] = (string)$alt['transcript'];
            if (!empty($alt['words']))      $allWords = array_merge($allWords, $alt['words']);
        }

        $transcript = implode(' ', $transcriptParts);
        if ($transcript === '') return null;

        // Group consecutive same-speaker words into segments
        $segments    = [];
        $speakerMap  = []; // raw speakerLabel → 'SPEAKER_XX'
        $curSegment  = null;

        foreach ($allWords as $word) {
            $rawLabel = (string)($word['speakerLabel'] ?? '');
            if ($rawLabel !== '' && !isset($speakerMap[$rawLabel])) {
                $speakerMap[$rawLabel] = sprintf('SPEAKER_%02d', count($speakerMap));
            }
            $speakerKey = $rawLabel !== '' ? $speakerMap[$rawLabel] : null;

            $start = isset($word['startOffset']) ? $this->offsetToSec((string)$word['startOffset']) : 0.0;
            $end   = isset($word['endOffset'])   ? $this->offsetToSec((string)$word['endOffset'])   : $start;
            $text  = (string)($word['word'] ?? '');

            $newSegment = $curSegment === null
                || ($speakerKey !== null && $speakerKey !== ($curSegment['speaker'] ?? null));

            if ($newSegment) {
                if ($curSegment !== null) $segments[] = $curSegment;
                $curSegment = ['text' => $text, 'start' => round($start, 3), 'end' => round($end, 3)];
                if ($speakerKey !== null) $curSegment['speaker'] = $speakerKey;
            } else {
                $curSegment['text'] .= ' ' . $text;
                $curSegment['end']   = round($end, 3);
            }
        }
        if ($curSegment !== null) $segments[] = $curSegment;

        // Duration from last word or last segment
        $durationSec = 0.0;
        if ($allWords) {
            $last = end($allWords);
            $durationSec = $this->offsetToSec((string)($last['endOffset'] ?? '0s'));
        } elseif ($segments) {
            $durationSec = (float)(end($segments)['end'] ?? 0.0);
        }

        return [
            'text'         => $transcript,
            'language'     => strtolower(explode('-', $locale)[0]),
            'duration'     => $durationSec,
            'segments'     => $segments,
            'num_speakers' => max(1, count($speakerMap)),
        ];
    }

    /** Convert GCP offset string like "1.200s" to float seconds. */
    private function offsetToSec(string $offset): float {
        return (float)rtrim($offset, 's');
    }

    private function resolveLocale(string $language): string {
        if ($language === '') return 'nb-NO';
        return self::LOCALE_MAP[$language] ?? $language;
    }
}