@mevdragon/vidfarm-devcli 0.10.0 → 0.12.0

package/dist/src/services/providers.js

@@ -5,35 +5,9 @@ import { createId } from "../lib/ids.js";
 import { addSeconds } from "../lib/time.js";
 import { ServerlessProviderKeyService } from "./serverless-provider-keys.js";
 import sharp from "sharp";
-export class ProviderKeyUnavailableError extends Error {
-}
-export class ProviderAuthError extends Error {
-}
-export class ProviderQuotaExceededError extends Error {
-}
-export class ProviderWaitExceededError extends Error {
-}
-export class ProviderLeaseTimeoutError extends Error {
-    retryAfterSeconds;
-    constructor(message, retryAfterSeconds = 5) {
-        super(message);
-        this.retryAfterSeconds = retryAfterSeconds;
-    }
-}
-export class ProviderRateLimitError extends Error {
-    retryAfterSeconds;
-    constructor(message, retryAfterSeconds = 60) {
-        super(message);
-        this.retryAfterSeconds = retryAfterSeconds;
-    }
-}
-export class ProviderTransientError extends Error {
-    retryAfterSeconds;
-    constructor(message, retryAfterSeconds = 5) {
-        super(message);
-        this.retryAfterSeconds = retryAfterSeconds;
-    }
-}
+import { ProviderAuthError, ProviderKeyUnavailableError, ProviderLeaseTimeoutError, ProviderQuotaExceededError, ProviderRateLimitError, ProviderTransientError, ProviderWaitExceededError, SpeechVoiceMismatchError, conciseProviderDetails, parseRetryAfterSeconds, rateLimitOrQuotaError } from "./provider-errors.js";
+import { SUPPORTED_STT_MODELS, SUPPORTED_TTS_MODELS, defaultSpeechModelFor, generateSpeechWithKey, inferSpeechContentType, speechVoiceFamilyFor, speechVoiceMismatch, transcribeSpeechWithKey } from "./speech.js";
+export { ProviderAuthError, ProviderKeyUnavailableError, ProviderLeaseTimeoutError, ProviderQuotaExceededError, ProviderRateLimitError, ProviderTransientError, ProviderWaitExceededError } from "./provider-errors.js";
 const SUPPORTED_TEXT_MODELS = {
     openai: new Set(["gpt-5.4"]),
     gemini: new Set(["gemini-3.1-flash-lite", "gemini-3.1-flash-lite-preview", "gemini-2.5-flash-lite"]),
@@ -96,16 +70,6 @@ const OPENAI_IMAGE_SIZES = new Set([
     "auto"
 ]);
 const GEMINI_IMAGE_ASPECT_RATIOS = new Set(["1:1", "16:9", "9:16", "4:3", "3:4", "21:9", "9:21"]);
-const SUPPORTED_TTS_MODELS = {
-    openrouter: new Set(["google/gemini-3.1-flash-tts-preview"]),
-    gemini: new Set(["gemini-3.1-flash-tts-preview", "gemini-2.5-flash-preview-tts"]),
-    openai: new Set(["gpt-4o-mini-tts-2025-12-15"])
-};
-const SUPPORTED_STT_MODELS = {
-    openrouter: new Set(["openai/gpt-4o-mini-transcribe"]),
-    gemini: new Set(["gemini-3.1-flash-lite-preview", "gemini-2.5-flash-lite"]),
-    openai: new Set(["gpt-4o-mini-transcribe-2025-12-15"])
-};
 const PROVIDER_KEY_VALIDATION_MODELS = {
     openai: "gpt-4.1-mini",
     gemini: "gemini-3.1-flash-lite-preview",
@@ -134,48 +98,9 @@ function formatDurationSeconds(totalSeconds) {
     }
     return `${Math.max(1, Math.round(totalSeconds))} second${Math.round(totalSeconds) === 1 ? "" : "s"}`;
 }
-function parseRetryAfterSeconds(value) {
-    if (!value) {
-        return 60;
-    }
-    const numeric = Number(value);
-    if (Number.isFinite(numeric) && numeric > 0) {
-        return Math.max(1, Math.min(Math.ceil(numeric), 60 * 60));
-    }
-    const parsedDate = Date.parse(value);
-    if (Number.isFinite(parsedDate)) {
-        return Math.max(1, Math.min(Math.ceil((parsedDate - Date.now()) / 1000), 60 * 60));
-    }
-    return 60;
-}
-function isHardQuotaError(details) {
-    const normalized = details.toLowerCase();
-    return [
-        "insufficient_quota",
-        "quota_exceeded",
-        "exceeded your current quota",
-        "billing",
-        "insufficient credits",
-        "not enough credits",
-        "credit balance",
-        "prepaid credits",
-        "resource_exhausted"
-    ].some((pattern) => normalized.includes(pattern));
-}
-function conciseProviderDetails(details) {
-    const compact = details.replace(/\s+/g, " ").trim();
-    return compact ? `: ${compact.slice(0, 500)}` : "";
-}
 function conciseProviderResponse(operation, response, details) {
     return `${operation} returned ${response.status}${conciseProviderDetails(details)}`;
 }
-async function rateLimitOrQuotaError(provider, response) {
-    const details = await response.text();
-    if (isHardQuotaError(details)) {
-        return new ProviderQuotaExceededError(`${provider} quota or billing limit was reached${conciseProviderDetails(details)}`);
-    }
-    return new ProviderRateLimitError(`${provider} rate limited${conciseProviderDetails(details)}`, parseRetryAfterSeconds(response.headers.get("retry-after")));
-}
 function isTransientProviderStatus(status) {
     return status === 408 || status === 409 || status === 425 || status === 500 || status === 502 || status === 503 || status === 504;
 }
@@ -253,9 +178,6 @@ function defaultModelFor(kind, provider) {
 function providerSupportsModelFrom(supported, provider, model) {
     return Boolean(supported[provider]?.has(normalizeProviderModel(model)));
 }
-function defaultSpeechModelFor(kind, provider) {
-    return kind === "tts" ? SUPPORTED_TTS_MODELS[provider]?.values().next().value : SUPPORTED_STT_MODELS[provider]?.values().next().value;
-}
 function uniqueProviders(requested, supported) {
     return [
         requested,
@@ -620,7 +542,9 @@ export class ProviderService {
         }
     }
     async generateSpeech(input) {
-        const attempts = await providerAttempts({
+        // Provider fallback must stay voice-aware: a Gemini preset routed to an
+        // OpenAI key (or the reverse) would 400 opaquely at the provider.
+        const attempts = (await providerAttempts({
             providerKeyService: this.serverlessProviderKeys,
             customerId: input.customerId,
             capability: "tts",
@@ -628,7 +552,11 @@ export class ProviderService {
             requestedModel: input.model,
             supportedModels: SUPPORTED_TTS_MODELS,
             defaultModelForProvider: (provider) => defaultSpeechModelFor("tts", provider)
-        });
+        })).filter((attempt) => !speechVoiceMismatch({ provider: attempt.provider, model: attempt.model, voice: input.voice }));
+        if (!attempts.length) {
+            throw new SpeechVoiceMismatchError(speechVoiceMismatch({ provider: input.provider, model: input.model, voice: input.voice }) ??
+                `Voice "${input.voice}" needs a ${speechVoiceFamilyFor(input.provider, input.model) === "gemini" ? "Gemini-family (gemini, or openrouter with a Gemini TTS model)" : "OpenAI-family (openai)"} provider key, but none is configured. Save a compatible key or pick a different voice.`);
+        }
         const retryDeadlineMs = providerLeaseRetryDeadlineMs();
         for (;;) {
             let lastError = null;
@@ -686,23 +614,15 @@ export class ProviderService {
                     }
                 };
             }
-            const result = zeroUsageCost(await this.withLeaseRenewal(lease, () => input.provider === "gemini"
-                ? this.callGeminiSpeechGeneration({
-                    model: input.model,
-                    text: input.text,
-                    voice: input.voice,
-                    instructions: input.instructions,
-                    apiKey: lease.secret
-                })
-                : this.callOpenAICompatibleSpeechGeneration({
-                    provider: input.provider === "openrouter" ? "openrouter" : "openai",
-                    model: input.model,
-                    text: input.text,
-                    voice: input.voice,
-                    instructions: input.instructions,
-                    responseFormat: input.responseFormat,
-                    apiKey: lease.secret
-                })));
+            const result = zeroUsageCost(await this.withLeaseRenewal(lease, () => generateSpeechWithKey({
+                provider: input.provider,
+                model: input.model,
+                text: input.text,
+                voice: input.voice,
+                instructions: input.instructions,
+                responseFormat: input.responseFormat,
+                apiKey: lease.secret
+            })));
             await this.serverlessProviderKeys.recordProviderKeyUsage({
                 id: createId("usage"),
                 keyId: lease.keyId,
@@ -1612,6 +1532,14 @@ export class ProviderService {
                 await this.serverlessProviderKeys.touchProviderKey(lease.keyId);
                 return {
                     text: "Mock transcription output.",
+                    language: input.language?.trim() || null,
+                    segments: input.diarize
+                        ? [
+                            { speaker: "speaker_1", start_sec: 0, end_sec: 2, text: "Mock transcription output." },
+                            { speaker: "speaker_2", start_sec: 2, end_sec: 4, text: "Mock second narrator." }
+                        ]
+                        : [{ speaker: "speaker_1", start_sec: null, end_sec: null, text: "Mock transcription output." }],
+                    diarization: input.diarize ? "diarized" : "none",
                     usage: {
                         inputTokens: 0,
                         outputTokens: 0,
@@ -1619,32 +1547,16 @@ export class ProviderService {
                     }
                 };
             }
-            const result = zeroUsageCost(await this.withLeaseRenewal(lease, () => input.provider === "gemini"
-                ? this.callGeminiSpeechTranscription({
-                    model: input.model,
-                    audio: input.audio,
-                    contentType: input.contentType,
-                    prompt: input.prompt,
-                    language: input.language,
-                    apiKey: lease.secret
-                })
-                : input.provider === "openrouter"
-                    ? this.callOpenRouterSpeechTranscription({
-                        model: input.model,
-                        audio: input.audio,
-                        contentType: input.contentType,
-                        prompt: input.prompt,
-                        language: input.language,
-                        apiKey: lease.secret
-                    })
-                    : this.callOpenAISpeechTranscription({
-                        model: input.model,
-                        audio: input.audio,
-                        contentType: input.contentType,
-                        prompt: input.prompt,
-                        language: input.language,
-                        apiKey: lease.secret
-                    })));
+            const result = zeroUsageCost(await this.withLeaseRenewal(lease, () => transcribeSpeechWithKey({
+                provider: input.provider,
+                model: input.model,
+                audio: input.audio,
+                contentType: input.contentType,
+                prompt: input.prompt,
+                language: input.language,
+                diarize: input.diarize,
+                apiKey: lease.secret
+            })));
             await this.serverlessProviderKeys.recordProviderKeyUsage({
                 id: createId("usage"),
                 keyId: lease.keyId,
@@ -1897,118 +1809,6 @@ export class ProviderService {
         }
         return image;
     }
-    async callOpenAICompatibleSpeechGeneration(input) {
-        const requestedFormat = input.responseFormat ?? "mp3";
-        const usePcmBridge = input.provider === "openrouter" &&
-            /gemini/i.test(input.model) &&
-            requestedFormat !== "mp3";
-        const responseFormat = usePcmBridge ? "pcm" : requestedFormat;
-        const endpoint = input.provider === "openrouter"
-            ? "https://openrouter.ai/api/v1/audio/speech"
-            : "https://api.openai.com/v1/audio/speech";
-        const response = await fetch(endpoint, {
-            method: "POST",
-            headers: {
-                "Content-Type": "application/json",
-                Authorization: `Bearer ${input.apiKey}`
-            },
-            body: JSON.stringify({
-                model: input.model,
-                input: input.text,
-                voice: input.voice ?? "alloy",
-                instructions: input.instructions,
-                response_format: responseFormat
-            })
-        });
-        if (response.status === 401 || response.status === 403) {
-            throw new ProviderAuthError(`${input.provider} authentication failed`);
-        }
-        if (response.status === 429) {
-            throw await rateLimitOrQuotaError(input.provider, response);
-        }
-        if (!response.ok) {
-            const details = await response.text();
-            throw new Error(`${input.provider} speech generation returned ${response.status}${details ? `: ${details}` : ""}`);
-        }
-        const rawBytes = Buffer.from(await response.arrayBuffer());
-        const bytes = usePcmBridge ? pcm16MonoToWav(rawBytes, 24000) : rawBytes;
-        const contentType = usePcmBridge
-            ? "audio/wav"
-            : response.headers.get("content-type")?.split(";")[0]?.trim() || inferSpeechContentType(responseFormat);
-        return {
-            bytes,
-            contentType,
-            usage: {
-                inputTokens: 0,
-                outputTokens: 0,
-                costUsd: 0
-            }
-        };
-    }
-    async callGeminiSpeechGeneration(input) {
-        const prompt = input.instructions?.trim()
-            ? `${input.instructions.trim()}\n\nSpeak exactly the following text:\n${input.text}`
-            : input.text;
-        const response = await fetch(`https://generativelanguage.googleapis.com/v1beta/models/${input.model}:generateContent?key=${input.apiKey}`, {
-            method: "POST",
-            headers: {
-                "Content-Type": "application/json"
-            },
-            body: JSON.stringify({
-                contents: [{ parts: [{ text: prompt }] }],
-                generationConfig: {
-                    responseModalities: ["AUDIO"],
-                    speechConfig: {
-                        voiceConfig: {
-                            prebuiltVoiceConfig: {
-                                voiceName: input.voice ?? "Kore"
-                            }
-                        }
-                    }
-                }
-            })
-        });
-        if (response.status === 401 || response.status === 403) {
-            throw new ProviderAuthError("gemini authentication failed");
-        }
-        if (response.status === 429) {
-            throw await rateLimitOrQuotaError("gemini", response);
-        }
-        if (!response.ok) {
-            const details = await response.text();
-            throw new Error(`gemini speech generation returned ${response.status}${details ? `: ${details}` : ""}`);
-        }
-        const data = await response.json();
-        const part = data.candidates?.[0]?.content?.parts?.find((item) => item.inlineData?.data || item.inline_data?.data);
-        const encoded = part?.inlineData?.data ?? part?.inline_data?.data;
-        if (!encoded) {
-            throw new Error("gemini speech generation returned no audio payload");
-        }
-        const rawBytes = Buffer.from(encoded, "base64");
-        const rawContentType = String(part?.inlineData?.mimeType ?? part?.inline_data?.mime_type ?? "audio/L16;rate=24000");
-        const pcmRateMatch = /audio\/l16;?\s*rate=(\d+)/i.exec(rawContentType);
-        const looksLikeWav = isWavBuffer(rawBytes);
-        const looksLikeMp3 = isLikelyMp3Buffer(rawBytes);
-        const shouldWrapPcm = Boolean(pcmRateMatch ||
-            (!looksLikeWav && !looksLikeMp3));
-        const bytes = shouldWrapPcm
-            ? pcm16MonoToWav(rawBytes, Number(pcmRateMatch?.[1] || "24000"))
-            : rawBytes;
-        const contentType = shouldWrapPcm
-            ? "audio/wav"
-            : looksLikeWav
-                ? "audio/wav"
-                : "audio/mpeg";
-        return {
-            bytes,
-            contentType,
-            usage: {
-                inputTokens: Number(data.usageMetadata?.promptTokenCount ?? 0),
-                outputTokens: Number(data.usageMetadata?.candidatesTokenCount ?? 0),
-                costUsd: 0
-            }
-        };
-    }
     async callOpenAICompatibleLayoutAnalysis(input) {
         const endpoint = input.provider === "openrouter"
             ? "https://openrouter.ai/api/v1/chat/completions"
@@ -2112,128 +1912,6 @@ export class ProviderService {
         const text = data.candidates?.[0]?.content?.parts?.map((part) => part.text ?? "").join("\n") ?? "";
         return String(text);
     }
-    async callOpenAISpeechTranscription(input) {
-        const contentType = input.contentType ?? "audio/mpeg";
-        const form = new FormData();
-        form.set("file", new Blob([Buffer.from(input.audio)], { type: contentType }), `audio.${audioExtensionFromContentType(contentType)}`);
-        form.set("model", input.model);
-        form.set("response_format", "json");
-        if (input.prompt) {
-            form.set("prompt", input.prompt);
-        }
-        if (input.language) {
-            form.set("language", input.language);
-        }
-        const response = await fetch("https://api.openai.com/v1/audio/transcriptions", {
-            method: "POST",
-            headers: {
-                Authorization: `Bearer ${input.apiKey}`
-            },
-            body: form
-        });
-        if (response.status === 401 || response.status === 403) {
-            throw new ProviderAuthError("openai authentication failed");
-        }
-        if (response.status === 429) {
-            throw await rateLimitOrQuotaError("openai", response);
-        }
-        if (!response.ok) {
-            const details = await response.text();
-            throw new Error(`openai transcription returned ${response.status}${details ? `: ${details}` : ""}`);
-        }
-        const data = await response.json();
-        return {
-            text: String(data.text ?? ""),
-            usage: {
-                inputTokens: Number(data.usage?.input_tokens ?? 0),
-                outputTokens: Number(data.usage?.output_tokens ?? 0),
-                costUsd: Number(data.usage?.cost ?? 0)
-            }
-        };
-    }
-    async callOpenRouterSpeechTranscription(input) {
-        const response = await fetch("https://openrouter.ai/api/v1/audio/transcriptions", {
-            method: "POST",
-            headers: {
-                "Content-Type": "application/json",
-                Authorization: `Bearer ${input.apiKey}`
-            },
-            body: JSON.stringify({
-                model: input.model,
-                input_audio: {
-                    data: Buffer.from(input.audio).toString("base64"),
-                    format: audioFormatFromContentType(input.contentType)
-                },
-                ...(input.prompt ? { prompt: input.prompt } : {}),
-                ...(input.language ? { language: input.language } : {})
-            })
-        });
-        if (response.status === 401 || response.status === 403) {
-            throw new ProviderAuthError("openrouter authentication failed");
-        }
-        if (response.status === 429) {
-            throw await rateLimitOrQuotaError("openrouter", response);
-        }
-        if (!response.ok) {
-            const details = await response.text();
-            throw new Error(`openrouter transcription returned ${response.status}${details ? `: ${details}` : ""}`);
-        }
-        const data = await response.json();
-        return {
-            text: String(data.text ?? ""),
-            usage: {
-                inputTokens: Number(data.usage?.input_tokens ?? data.usage?.prompt_tokens ?? 0),
-                outputTokens: Number(data.usage?.output_tokens ?? data.usage?.completion_tokens ?? 0),
-                costUsd: Number(data.usage?.cost ?? 0)
-            }
-        };
-    }
-    async callGeminiSpeechTranscription(input) {
-        const response = await fetch(`https://generativelanguage.googleapis.com/v1beta/models/${input.model}:generateContent?key=${input.apiKey}`, {
-            method: "POST",
-            headers: {
-                "Content-Type": "application/json"
-            },
-            body: JSON.stringify({
-                contents: [{
-                        parts: [
-                            {
-                                inline_data: {
-                                    mime_type: input.contentType ?? "audio/mpeg",
-                                    data: Buffer.from(input.audio).toString("base64")
-                                }
-                            },
-                            {
-                                text: buildGeminiTranscriptionPrompt(input.prompt, input.language)
-                            }
-                        ]
-                    }],
-                generationConfig: {
-                    temperature: 0,
-                    responseMimeType: "text/plain"
-                }
-            })
-        });
-        if (response.status === 401 || response.status === 403) {
-            throw new ProviderAuthError("gemini authentication failed");
-        }
-        if (response.status === 429) {
-            throw await rateLimitOrQuotaError("gemini", response);
-        }
-        if (!response.ok) {
-            const details = await response.text();
-            throw new Error(`gemini transcription returned ${response.status}${details ? `: ${details}` : ""}`);
-        }
-        const data = await response.json();
-        return {
-            text: extractGeminiText(data),
-            usage: {
-                inputTokens: Number(data.usageMetadata?.promptTokenCount ?? 0),
-                outputTokens: Number(data.usageMetadata?.candidatesTokenCount ?? 0),
-                costUsd: 0
-            }
-        };
-    }
     async callProvider(input) {
         if (input.provider === "gemini") {
             return this.callGemini({
@@ -2643,77 +2321,6 @@ function assertSupportedSpeechModel(kind, provider, model) {
 function normalizeProviderModel(model) {
     return model.trim().replace(/^models\//, "");
 }
-function inferSpeechContentType(format) {
-    return format === "wav" ? "audio/wav" : "audio/mpeg";
-}
-function pcm16MonoToWav(pcmBuffer, sampleRate) {
-    const header = Buffer.alloc(44);
-    header.write("RIFF", 0, 4, "ascii");
-    header.writeUInt32LE(36 + pcmBuffer.byteLength, 4);
-    header.write("WAVE", 8, 4, "ascii");
-    header.write("fmt ", 12, 4, "ascii");
-    header.writeUInt32LE(16, 16);
-    header.writeUInt16LE(1, 20);
-    header.writeUInt16LE(1, 22);
-    header.writeUInt32LE(sampleRate, 24);
-    header.writeUInt32LE(sampleRate * 2, 28);
-    header.writeUInt16LE(2, 32);
-    header.writeUInt16LE(16, 34);
-    header.write("data", 36, 4, "ascii");
-    header.writeUInt32LE(pcmBuffer.byteLength, 40);
-    return Buffer.concat([header, pcmBuffer]);
-}
-function isWavBuffer(buffer) {
-    return buffer.byteLength >= 12
-        && buffer.subarray(0, 4).toString("ascii") === "RIFF"
-        && buffer.subarray(8, 12).toString("ascii") === "WAVE";
-}
-function isLikelyMp3Buffer(buffer) {
-    if (buffer.byteLength < 3) {
-        return false;
-    }
-    if (buffer.subarray(0, 3).toString("ascii") === "ID3") {
-        return true;
-    }
-    return buffer[0] === 0xff && (buffer[1] & 0xe0) === 0xe0;
-}
-function audioFormatFromContentType(contentType) {
-    const normalized = contentType?.split(";")[0]?.trim().toLowerCase();
-    switch (normalized) {
-        case "audio/wav":
-        case "audio/x-wav":
-            return "wav";
-        case "audio/mp4":
-        case "audio/m4a":
-            return "mp4";
-        case "audio/webm":
-            return "webm";
-        case "audio/ogg":
-            return "ogg";
-        default:
-            return "mp3";
-    }
-}
-function audioExtensionFromContentType(contentType) {
-    switch (audioFormatFromContentType(contentType)) {
-        case "wav":
-            return "wav";
-        case "mp4":
-            return "m4a";
-        case "webm":
-            return "webm";
-        case "ogg":
-            return "ogg";
-        default:
-            return "mp3";
-    }
-}
-function buildGeminiTranscriptionPrompt(prompt, language) {
-    const basePrompt = prompt?.trim() || "Transcribe this audio verbatim. Return plain text only.";
-    return language?.trim()
-        ? `${basePrompt}\n\nExpected language: ${language.trim()}.`
-        : basePrompt;
-}
 function extractGeminiText(data) {
     return String(data.candidates?.[0]?.content?.parts?.map((part) => part.text ?? "").join("\n") ?? "");
 }