@damiangilgonzalez/copypasta 0.0.0-stage → 0.1.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +227 -2
- package/catalog.yaml +140 -0
- package/cli/copypasta.js +4 -0
- package/cli/lib/catalog.js +131 -0
- package/cli/lib/commands.js +162 -0
- package/cli/lib/index.js +4 -0
- package/cli/lib/main.js +130 -0
- package/package.json +38 -4
- package/recipes/agents/agent-memory/RECIPE.md +75 -0
- package/recipes/agents/agent-memory/files/agentic/memory/__init__.py +25 -0
- package/recipes/agents/agent-memory/files/agentic/memory/base.py +101 -0
- package/recipes/agents/agent-memory/files/agentic/memory/factory.py +33 -0
- package/recipes/agents/agent-memory/files/agentic/memory/in_memory.py +59 -0
- package/recipes/agents/agent-memory/files/agentic/memory/models.py +66 -0
- package/recipes/agents/agent-memory/files/agentic/memory/settings.py +24 -0
- package/recipes/agents/agent-memory/files/agentic/memory/sql.py +177 -0
- package/recipes/agents/agent-memory/files/pytest.ini +5 -0
- package/recipes/agents/agent-memory/files/tests/__init__.py +0 -0
- package/recipes/agents/agent-memory/files/tests/memory_conformance.py +158 -0
- package/recipes/agents/agent-memory/files/tests/test_memory.py +74 -0
- package/recipes/agents/langchain-agent/RECIPE.md +101 -0
- package/recipes/agents/langchain-agent/files/agentic/__init__.py +10 -0
- package/recipes/agents/langchain-agent/files/agentic/agents/__init__.py +13 -0
- package/recipes/agents/langchain-agent/files/agentic/agents/agents.py +43 -0
- package/recipes/agents/langchain-agent/files/agentic/agents/base_agent.py +126 -0
- package/recipes/agents/langchain-agent/files/agentic/agents/models.py +46 -0
- package/recipes/agents/langchain-agent/files/agentic/context.py +21 -0
- package/recipes/agents/langchain-agent/files/agentic/llm/__init__.py +5 -0
- package/recipes/agents/langchain-agent/files/agentic/llm/models.py +45 -0
- package/recipes/agents/langchain-agent/files/agentic/llm/provider.py +78 -0
- package/recipes/agents/langchain-agent/files/agentic/llm/settings.py +62 -0
- package/recipes/agents/langchain-agent/files/agentic/middleware/__init__.py +3 -0
- package/recipes/agents/langchain-agent/files/agentic/middleware/language.py +28 -0
- package/recipes/agents/langchain-agent/files/agentic/prompts/__init__.py +3 -0
- package/recipes/agents/langchain-agent/files/agentic/prompts/assistant.en.yaml +15 -0
- package/recipes/agents/langchain-agent/files/agentic/prompts/assistant.es.yaml +15 -0
- package/recipes/agents/langchain-agent/files/agentic/prompts/loader.py +69 -0
- package/recipes/agents/langchain-agent/files/agentic/tools/__init__.py +25 -0
- package/recipes/agents/langchain-agent/files/agentic/tools/base_tool.py +62 -0
- package/recipes/agents/langchain-agent/files/agentic/tools/search_documents.py +42 -0
- package/recipes/agents/langchain-agent/files/pytest.ini +5 -0
- package/recipes/agents/langchain-agent/files/tests/__init__.py +0 -0
- package/recipes/agents/langchain-agent/files/tests/fakes.py +30 -0
- package/recipes/agents/langchain-agent/files/tests/test_agents.py +151 -0
- package/recipes/agents/langgraph-graph/RECIPE.md +63 -0
- package/recipes/agents/langgraph-graph/files/agentic/conversation.py +73 -0
- package/recipes/agents/langgraph-graph/files/agentic/graph/__init__.py +17 -0
- package/recipes/agents/langgraph-graph/files/agentic/graph/builder.py +37 -0
- package/recipes/agents/langgraph-graph/files/agentic/graph/edges.py +8 -0
- package/recipes/agents/langgraph-graph/files/agentic/graph/event_processor.py +48 -0
- package/recipes/agents/langgraph-graph/files/agentic/graph/events.py +30 -0
- package/recipes/agents/langgraph-graph/files/agentic/graph/graph.py +60 -0
- package/recipes/agents/langgraph-graph/files/agentic/graph/nodes.py +63 -0
- package/recipes/agents/langgraph-graph/files/agentic/graph/state.py +31 -0
- package/recipes/agents/langgraph-graph/files/agentic/registry.py +85 -0
- package/recipes/agents/langgraph-graph/files/tests/test_graph.py +145 -0
- package/recipes/backend/.gitkeep +0 -0
- package/recipes/backend/agent-chat-api/RECIPE.md +69 -0
- package/recipes/backend/agent-chat-api/files/api/agent/guards.py +35 -0
- package/recipes/backend/agent-chat-api/files/api/agent/router.py +86 -0
- package/recipes/backend/agent-chat-api/files/api/agent/sanitize.py +22 -0
- package/recipes/backend/agent-chat-api/files/api/agent/schemas.py +26 -0
- package/recipes/backend/agent-chat-api/files/api/agent/settings.py +14 -0
- package/recipes/backend/agent-chat-api/files/api/agent/tracing.py +38 -0
- package/recipes/backend/agent-chat-api/files/api/app.py +67 -0
- package/recipes/backend/agent-chat-api/files/api/auth.py +73 -0
- package/recipes/backend/agent-chat-api/files/pytest.ini +5 -0
- package/recipes/backend/agent-chat-api/files/tests/test_agent_api.py +237 -0
- package/recipes/backend/agent-feedback/RECIPE.md +63 -0
- package/recipes/backend/agent-feedback/files/api/feedback/__init__.py +0 -0
- package/recipes/backend/agent-feedback/files/api/feedback/router.py +72 -0
- package/recipes/backend/agent-feedback/files/api/feedback/schemas.py +44 -0
- package/recipes/backend/agent-feedback/files/feedback/__init__.py +25 -0
- package/recipes/backend/agent-feedback/files/feedback/base.py +53 -0
- package/recipes/backend/agent-feedback/files/feedback/factory.py +30 -0
- package/recipes/backend/agent-feedback/files/feedback/in_memory.py +29 -0
- package/recipes/backend/agent-feedback/files/feedback/models.py +48 -0
- package/recipes/backend/agent-feedback/files/feedback/settings.py +17 -0
- package/recipes/backend/agent-feedback/files/feedback/sql.py +102 -0
- package/recipes/backend/agent-feedback/files/pytest.ini +5 -0
- package/recipes/backend/agent-feedback/files/tests/feedback_conformance.py +89 -0
- package/recipes/backend/agent-feedback/files/tests/test_feedback.py +63 -0
- package/recipes/backend/agent-feedback/files/tests/test_feedback_api.py +95 -0
- package/recipes/backend/auth-verifier/RECIPE.md +63 -0
- package/recipes/backend/auth-verifier/files/api/auth.py +73 -0
- package/recipes/backend/auth-verifier/files/auth/__init__.py +23 -0
- package/recipes/backend/auth-verifier/files/auth/factory.py +28 -0
- package/recipes/backend/auth-verifier/files/auth/jwt_verifier.py +101 -0
- package/recipes/backend/auth-verifier/files/auth/models.py +29 -0
- package/recipes/backend/auth-verifier/files/auth/settings.py +45 -0
- package/recipes/backend/auth-verifier/files/auth/supabase.py +80 -0
- package/recipes/backend/auth-verifier/files/pytest.ini +5 -0
- package/recipes/backend/auth-verifier/files/tests/__init__.py +0 -0
- package/recipes/backend/auth-verifier/files/tests/test_auth.py +229 -0
- package/recipes/backend/conversation-history/RECIPE.md +49 -0
- package/recipes/backend/conversation-history/files/api/conversations/__init__.py +0 -0
- package/recipes/backend/conversation-history/files/api/conversations/router.py +55 -0
- package/recipes/backend/conversation-history/files/api/conversations/schemas.py +57 -0
- package/recipes/backend/conversation-history/files/pytest.ini +5 -0
- package/recipes/backend/conversation-history/files/tests/test_conversations_api.py +127 -0
- package/recipes/backend/fastapi-base/RECIPE.md +57 -0
- package/recipes/backend/fastapi-base/files/api/app.py +37 -0
- package/recipes/backend/fastapi-base/files/api/base.py +54 -0
- package/recipes/backend/fastapi-base/files/api/errors.py +31 -0
- package/recipes/backend/fastapi-base/files/api/health.py +19 -0
- package/recipes/backend/fastapi-base/files/api/logging_setup.py +26 -0
- package/recipes/backend/fastapi-base/files/api/request_id.py +62 -0
- package/recipes/backend/fastapi-base/files/api/settings.py +32 -0
- package/recipes/backend/fastapi-base/files/pytest.ini +5 -0
- package/recipes/backend/fastapi-base/files/tests/__init__.py +0 -0
- package/recipes/backend/fastapi-base/files/tests/test_api_base.py +136 -0
- package/recipes/backend/observability/RECIPE.md +68 -0
- package/recipes/backend/observability/files/observability/__init__.py +34 -0
- package/recipes/backend/observability/files/observability/factory.py +73 -0
- package/recipes/backend/observability/files/observability/langfuse_tracer.py +29 -0
- package/recipes/backend/observability/files/observability/langsmith_tracer.py +23 -0
- package/recipes/backend/observability/files/observability/models.py +90 -0
- package/recipes/backend/observability/files/observability/opik_tracer.py +30 -0
- package/recipes/backend/observability/files/observability/settings.py +58 -0
- package/recipes/backend/observability/files/observability/tracer.py +142 -0
- package/recipes/backend/observability/files/observability/usage.py +61 -0
- package/recipes/backend/observability/files/pytest.ini +5 -0
- package/recipes/backend/observability/files/tests/__init__.py +0 -0
- package/recipes/backend/observability/files/tests/test_observability.py +219 -0
- package/recipes/backend/project-settings/RECIPE.md +40 -0
- package/recipes/backend/project-settings/files/.env.example +7 -0
- package/recipes/backend/project-settings/files/core/__init__.py +0 -0
- package/recipes/backend/project-settings/files/core/config.py +41 -0
- package/recipes/backend/project-settings/files/pytest.ini +3 -0
- package/recipes/backend/project-settings/files/tests/__init__.py +0 -0
- package/recipes/backend/project-settings/files/tests/test_settings.py +44 -0
- package/recipes/backend/rag-retriever/RECIPE.md +79 -0
- package/recipes/backend/rag-retriever/files/pytest.ini +5 -0
- package/recipes/backend/rag-retriever/files/rag/__init__.py +53 -0
- package/recipes/backend/rag-retriever/files/rag/base.py +106 -0
- package/recipes/backend/rag-retriever/files/rag/chunking.py +84 -0
- package/recipes/backend/rag-retriever/files/rag/document_search.py +60 -0
- package/recipes/backend/rag-retriever/files/rag/embeddings.py +91 -0
- package/recipes/backend/rag-retriever/files/rag/factory.py +82 -0
- package/recipes/backend/rag-retriever/files/rag/in_memory.py +52 -0
- package/recipes/backend/rag-retriever/files/rag/models.py +107 -0
- package/recipes/backend/rag-retriever/files/rag/pgvector.py +138 -0
- package/recipes/backend/rag-retriever/files/rag/qdrant.py +96 -0
- package/recipes/backend/rag-retriever/files/rag/settings.py +48 -0
- package/recipes/backend/rag-retriever/files/tests/__init__.py +0 -0
- package/recipes/backend/rag-retriever/files/tests/rag_conformance.py +95 -0
- package/recipes/backend/rag-retriever/files/tests/test_rag.py +270 -0
- package/recipes/backend/rate-limit-quota/RECIPE.md +77 -0
- package/recipes/backend/rate-limit-quota/files/api/agent/guards.py +62 -0
- package/recipes/backend/rate-limit-quota/files/api/limits/__init__.py +0 -0
- package/recipes/backend/rate-limit-quota/files/api/limits/router.py +38 -0
- package/recipes/backend/rate-limit-quota/files/limits/__init__.py +43 -0
- package/recipes/backend/rate-limit-quota/files/limits/base.py +161 -0
- package/recipes/backend/rate-limit-quota/files/limits/factory.py +69 -0
- package/recipes/backend/rate-limit-quota/files/limits/in_memory.py +36 -0
- package/recipes/backend/rate-limit-quota/files/limits/models.py +108 -0
- package/recipes/backend/rate-limit-quota/files/limits/settings.py +32 -0
- package/recipes/backend/rate-limit-quota/files/limits/sql.py +94 -0
- package/recipes/backend/rate-limit-quota/files/pytest.ini +5 -0
- package/recipes/backend/rate-limit-quota/files/tests/limits_conformance.py +146 -0
- package/recipes/backend/rate-limit-quota/files/tests/test_limits.py +167 -0
- package/recipes/backend/rate-limit-quota/files/tests/test_limits_api.py +100 -0
- package/recipes/backend/voice-providers/RECIPE.md +73 -0
- package/recipes/backend/voice-providers/files/api/__init__.py +0 -0
- package/recipes/backend/voice-providers/files/api/voice/__init__.py +0 -0
- package/recipes/backend/voice-providers/files/api/voice/router.py +113 -0
- package/recipes/backend/voice-providers/files/pytest.ini +5 -0
- package/recipes/backend/voice-providers/files/tests/__init__.py +0 -0
- package/recipes/backend/voice-providers/files/tests/fake_openai.py +44 -0
- package/recipes/backend/voice-providers/files/tests/test_voice.py +164 -0
- package/recipes/backend/voice-providers/files/tests/test_voice_api.py +106 -0
- package/recipes/backend/voice-providers/files/tests/voice_conformance.py +80 -0
- package/recipes/backend/voice-providers/files/voice/__init__.py +29 -0
- package/recipes/backend/voice-providers/files/voice/base.py +116 -0
- package/recipes/backend/voice-providers/files/voice/factory.py +48 -0
- package/recipes/backend/voice-providers/files/voice/fake.py +51 -0
- package/recipes/backend/voice-providers/files/voice/models.py +85 -0
- package/recipes/backend/voice-providers/files/voice/openai_provider.py +116 -0
- package/recipes/backend/voice-providers/files/voice/settings.py +48 -0
- package/recipes/frontend/.gitkeep +0 -0
- package/recipes/frontend/playwright-e2e/RECIPE.md +48 -0
- package/recipes/frontend/playwright-e2e/files/.github/workflows/frontend.yml +35 -0
- package/recipes/frontend/playwright-e2e/files/e2e/helpers.ts +25 -0
- package/recipes/frontend/playwright-e2e/files/e2e/home.spec.ts +10 -0
- package/recipes/frontend/playwright-e2e/files/e2e/login.spec.ts +26 -0
- package/recipes/frontend/playwright-e2e/files/e2e/navigation.spec.ts +27 -0
- package/recipes/frontend/playwright-e2e/files/playwright.config.ts +35 -0
- package/recipes/frontend/react-agent-chat/RECIPE.md +69 -0
- package/recipes/frontend/react-agent-chat/files/src/features/chat/chat-api.ts +27 -0
- package/recipes/frontend/react-agent-chat/files/src/features/chat/chat.css +18 -0
- package/recipes/frontend/react-agent-chat/files/src/features/chat/chat.test.tsx +139 -0
- package/recipes/frontend/react-agent-chat/files/src/features/chat/components/ChatView.tsx +22 -0
- package/recipes/frontend/react-agent-chat/files/src/features/chat/components/Composer.tsx +40 -0
- package/recipes/frontend/react-agent-chat/files/src/features/chat/components/Markdown.tsx +20 -0
- package/recipes/frontend/react-agent-chat/files/src/features/chat/components/MessageList.tsx +42 -0
- package/recipes/frontend/react-agent-chat/files/src/features/chat/events.ts +42 -0
- package/recipes/frontend/react-agent-chat/files/src/features/chat/fake-chat-api.ts +44 -0
- package/recipes/frontend/react-agent-chat/files/src/features/chat/index.ts +6 -0
- package/recipes/frontend/react-agent-chat/files/src/features/chat/sse-parser.test.ts +97 -0
- package/recipes/frontend/react-agent-chat/files/src/features/chat/sse-parser.ts +85 -0
- package/recipes/frontend/react-agent-chat/files/src/features/chat/use-chat.ts +136 -0
- package/recipes/frontend/react-app-router/RECIPE.md +58 -0
- package/recipes/frontend/react-app-router/files/src/App.test.tsx +54 -0
- package/recipes/frontend/react-app-router/files/src/App.tsx +14 -0
- package/recipes/frontend/react-app-router/files/src/router/ErrorScreen.tsx +21 -0
- package/recipes/frontend/react-app-router/files/src/router/Layout.tsx +28 -0
- package/recipes/frontend/react-app-router/files/src/router/NotFoundScreen.tsx +12 -0
- package/recipes/frontend/react-app-router/files/src/router/index.ts +1 -0
- package/recipes/frontend/react-app-router/files/src/router/router.css +29 -0
- package/recipes/frontend/react-app-router/files/src/router/router.test.tsx +65 -0
- package/recipes/frontend/react-app-router/files/src/router/routes.tsx +59 -0
- package/recipes/frontend/react-app-router/files/src/screens/SettingsScreen.tsx +10 -0
- package/recipes/frontend/react-supabase-auth/RECIPE.md +77 -0
- package/recipes/frontend/react-supabase-auth/files/src/auth/LoginScreen.tsx +64 -0
- package/recipes/frontend/react-supabase-auth/files/src/auth/SignOutButton.tsx +12 -0
- package/recipes/frontend/react-supabase-auth/files/src/auth/auth.css +33 -0
- package/recipes/frontend/react-supabase-auth/files/src/auth/auth.test.ts +149 -0
- package/recipes/frontend/react-supabase-auth/files/src/auth/auth.ts +59 -0
- package/recipes/frontend/react-supabase-auth/files/src/auth/fake-auth.ts +87 -0
- package/recipes/frontend/react-supabase-auth/files/src/auth/fake-gotrue.ts +85 -0
- package/recipes/frontend/react-supabase-auth/files/src/auth/index.ts +7 -0
- package/recipes/frontend/react-supabase-auth/files/src/auth/make-auth.ts +20 -0
- package/recipes/frontend/react-supabase-auth/files/src/auth/session-context.tsx +35 -0
- package/recipes/frontend/react-supabase-auth/files/src/auth/session-store.ts +63 -0
- package/recipes/frontend/react-supabase-auth/files/src/auth/session.test.tsx +56 -0
- package/recipes/frontend/react-supabase-auth/files/src/auth/settings.ts +30 -0
- package/recipes/frontend/react-supabase-auth/files/src/auth/supabase-auth.ts +85 -0
- package/recipes/frontend/vite-react-base/RECIPE.md +69 -0
- package/recipes/frontend/vite-react-base/files/.env +6 -0
- package/recipes/frontend/vite-react-base/files/.env.fake +2 -0
- package/recipes/frontend/vite-react-base/files/index.html +12 -0
- package/recipes/frontend/vite-react-base/files/package.json +30 -0
- package/recipes/frontend/vite-react-base/files/src/App.test.tsx +49 -0
- package/recipes/frontend/vite-react-base/files/src/App.tsx +13 -0
- package/recipes/frontend/vite-react-base/files/src/api/api.ts +37 -0
- package/recipes/frontend/vite-react-base/files/src/api/fake/fake-api.ts +29 -0
- package/recipes/frontend/vite-react-base/files/src/api/http-api.ts +16 -0
- package/recipes/frontend/vite-react-base/files/src/api/http-client.test.ts +84 -0
- package/recipes/frontend/vite-react-base/files/src/api/http-client.ts +94 -0
- package/recipes/frontend/vite-react-base/files/src/api/make-api.ts +30 -0
- package/recipes/frontend/vite-react-base/files/src/api/types.ts +7 -0
- package/recipes/frontend/vite-react-base/files/src/app-context.tsx +24 -0
- package/recipes/frontend/vite-react-base/files/src/auth/token-provider.ts +27 -0
- package/recipes/frontend/vite-react-base/files/src/config.ts +18 -0
- package/recipes/frontend/vite-react-base/files/src/hooks/use-load.ts +46 -0
- package/recipes/frontend/vite-react-base/files/src/main.tsx +27 -0
- package/recipes/frontend/vite-react-base/files/src/screens/HomeScreen.tsx +28 -0
- package/recipes/frontend/vite-react-base/files/src/styles/global.css +25 -0
- package/recipes/frontend/vite-react-base/files/src/styles/tokens.css +33 -0
- package/recipes/frontend/vite-react-base/files/src/styles/tokens.test.ts +45 -0
- package/recipes/frontend/vite-react-base/files/src/test-setup.ts +8 -0
- package/recipes/frontend/vite-react-base/files/src/vite-env.d.ts +11 -0
- package/recipes/frontend/vite-react-base/files/tsconfig.json +18 -0
- package/recipes/frontend/vite-react-base/files/vite.config.ts +19 -0
- package/unpacked-files.json +3 -0
|
@@ -0,0 +1,164 @@
|
|
|
1
|
+
import wave
|
|
2
|
+
import io
|
|
3
|
+
|
|
4
|
+
import pytest
|
|
5
|
+
from pydantic import SecretStr
|
|
6
|
+
|
|
7
|
+
from voice import AudioInput, SttProvider, TtsProvider, VoiceError, VoiceErrorCode, VoiceFactory, VoiceSettings
|
|
8
|
+
from voice.fake import FakeSynthesizer, FakeTranscriber
|
|
9
|
+
from voice.openai_provider import OpenAISynthesizer, OpenAITranscriber, translate_openai_error
|
|
10
|
+
|
|
11
|
+
from .fake_openai import FakeOpenAI
|
|
12
|
+
from .voice_conformance import SILENCE, WEBM, SynthesizerConformance, TranscriberConformance
|
|
13
|
+
|
|
14
|
+
|
|
15
|
+
def settings(**overrides) -> VoiceSettings:
|
|
16
|
+
base = {"stt_provider": "fake", "tts_provider": "fake"}
|
|
17
|
+
return VoiceSettings(_env_file=None, **{**base, **overrides})
|
|
18
|
+
|
|
19
|
+
|
|
20
|
+
def openai_settings(fake: FakeOpenAI, **overrides) -> VoiceSettings:
|
|
21
|
+
return settings(stt_provider="openai", tts_provider="openai", openai_api_key="sk-test", **overrides)
|
|
22
|
+
|
|
23
|
+
|
|
24
|
+
# --- providers through the conformance batteries -------------------------------------------
|
|
25
|
+
|
|
26
|
+
|
|
27
|
+
class TestFakeTranscriber(TranscriberConformance):
|
|
28
|
+
@pytest.fixture
|
|
29
|
+
def transcriber(self):
|
|
30
|
+
return FakeTranscriber(text="hola mundo", language="es")
|
|
31
|
+
|
|
32
|
+
|
|
33
|
+
class TestOpenAITranscriber(TranscriberConformance):
|
|
34
|
+
@pytest.fixture
|
|
35
|
+
def transcriber(self):
|
|
36
|
+
fake = FakeOpenAI(text="hola mundo")
|
|
37
|
+
|
|
38
|
+
class Silent(FakeOpenAI):
|
|
39
|
+
def _handle(self, request):
|
|
40
|
+
if b"SILENCE" in request.read():
|
|
41
|
+
self.text = ""
|
|
42
|
+
return super()._handle(request)
|
|
43
|
+
|
|
44
|
+
return VoiceFactory.transcriber(openai_settings(fake), http_client=Silent(text="hola mundo").http_client)
|
|
45
|
+
|
|
46
|
+
|
|
47
|
+
class TestFakeSynthesizer(SynthesizerConformance):
|
|
48
|
+
@pytest.fixture
|
|
49
|
+
def synthesizer(self):
|
|
50
|
+
return FakeSynthesizer()
|
|
51
|
+
|
|
52
|
+
|
|
53
|
+
class TestOpenAISynthesizer(SynthesizerConformance):
|
|
54
|
+
@pytest.fixture
|
|
55
|
+
def synthesizer(self):
|
|
56
|
+
return VoiceFactory.synthesizer(openai_settings(FakeOpenAI()), http_client=FakeOpenAI().http_client)
|
|
57
|
+
|
|
58
|
+
|
|
59
|
+
# --- what is sent to OpenAI ------------------------------------------------------------------
|
|
60
|
+
|
|
61
|
+
|
|
62
|
+
async def test_openai_transcriber_sends_file_model_language_and_prompt():
|
|
63
|
+
fake = FakeOpenAI(text=" hola mundo ")
|
|
64
|
+
transcriber = VoiceFactory.transcriber(
|
|
65
|
+
openai_settings(fake, stt_model="gpt-4o-mini-transcribe", stt_language="es", stt_prompt="Acme, SKU"), http_client=fake.http_client
|
|
66
|
+
)
|
|
67
|
+
result = await transcriber.transcribe(AudioInput(content=b"...", mime="audio/mp4; codecs=mp4a"), language="pt")
|
|
68
|
+
assert result.text == "hola mundo" and result.language == "pt" and result.engine == "openai"
|
|
69
|
+
sent = fake.requests[0]
|
|
70
|
+
assert sent["path"].endswith("/audio/transcriptions") and sent["filename"] == "audio.mp4"
|
|
71
|
+
assert sent["fields"]["model"] == "gpt-4o-mini-transcribe" and sent["fields"]["language"] == "pt"
|
|
72
|
+
assert sent["fields"]["prompt"] == "Acme, SKU" and sent["fields"]["response_format"] == "json"
|
|
73
|
+
|
|
74
|
+
|
|
75
|
+
async def test_whisper_uses_verbose_json_and_drops_hallucinations_on_silence():
|
|
76
|
+
noisy = FakeOpenAI(text="Subtítulos por Amara.org", segments=[{"no_speech_prob": 0.95, "text": "x"}, {"no_speech_prob": 0.7, "text": "y"}])
|
|
77
|
+
transcriber = VoiceFactory.transcriber(openai_settings(noisy, stt_model="whisper-1"), http_client=noisy.http_client)
|
|
78
|
+
assert (await transcriber.transcribe(WEBM)).text == ""
|
|
79
|
+
assert noisy.requests[0]["fields"]["response_format"] == "verbose_json"
|
|
80
|
+
speech = FakeOpenAI(text="hola", segments=[{"no_speech_prob": 0.95, "text": "x"}, {"no_speech_prob": 0.1, "text": "hola"}])
|
|
81
|
+
transcriber = VoiceFactory.transcriber(openai_settings(speech, stt_model="whisper-1"), http_client=speech.http_client)
|
|
82
|
+
assert (await transcriber.transcribe(WEBM)).text == "hola"
|
|
83
|
+
|
|
84
|
+
|
|
85
|
+
async def test_openai_synthesizer_sends_model_voice_and_format():
|
|
86
|
+
fake = FakeOpenAI()
|
|
87
|
+
synthesizer = VoiceFactory.synthesizer(openai_settings(fake, tts_model="tts-1-hd", tts_voice="nova", tts_format="mp3"), http_client=fake.http_client)
|
|
88
|
+
speech = await synthesizer.synthesize("Hola", voice="echo")
|
|
89
|
+
assert speech.audio.startswith(b"ID3") and speech.mime == "audio/mpeg"
|
|
90
|
+
assert fake.requests[0]["json"] == {"model": "tts-1-hd", "voice": "echo", "input": "Hola", "response_format": "mp3"}
|
|
91
|
+
|
|
92
|
+
|
|
93
|
+
@pytest.mark.parametrize(
|
|
94
|
+
"status, body, code, retryable",
|
|
95
|
+
[
|
|
96
|
+
(429, None, VoiceErrorCode.RATE_LIMITED, True),
|
|
97
|
+
(413, None, VoiceErrorCode.INVALID_AUDIO, False),
|
|
98
|
+
(415, None, VoiceErrorCode.INVALID_AUDIO, False),
|
|
99
|
+
(400, {"error": {"message": "audio too short", "param": "file", "type": "x", "code": "audio_too_short"}}, VoiceErrorCode.INVALID_AUDIO, False),
|
|
100
|
+
(400, {"error": {"message": "bad model", "param": "model", "type": "x", "code": None}}, VoiceErrorCode.UNAVAILABLE, False),
|
|
101
|
+
(401, None, VoiceErrorCode.UNAVAILABLE, False),
|
|
102
|
+
(500, None, VoiceErrorCode.UNAVAILABLE, True),
|
|
103
|
+
(503, None, VoiceErrorCode.UNAVAILABLE, True),
|
|
104
|
+
],
|
|
105
|
+
)
|
|
106
|
+
async def test_openai_http_errors_become_voice_errors_without_the_message(status, body, code, retryable, caplog):
|
|
107
|
+
fake = FakeOpenAI(status=status, body=body)
|
|
108
|
+
transcriber = VoiceFactory.transcriber(openai_settings(fake), http_client=fake.http_client)
|
|
109
|
+
with pytest.raises(VoiceError) as info:
|
|
110
|
+
await transcriber.transcribe(WEBM)
|
|
111
|
+
assert (info.value.code, info.value.retryable) == (code, retryable)
|
|
112
|
+
assert "nope" not in str(info.value) and "nope" not in caplog.text and info.value.kind.endswith("Error")
|
|
113
|
+
|
|
114
|
+
|
|
115
|
+
def test_translate_connection_and_timeout_errors():
|
|
116
|
+
import httpx
|
|
117
|
+
import openai
|
|
118
|
+
|
|
119
|
+
request = httpx.Request("POST", "http://x")
|
|
120
|
+
assert translate_openai_error(openai.APITimeoutError(request)).code is VoiceErrorCode.TIMEOUT
|
|
121
|
+
connection = translate_openai_error(openai.APIConnectionError(request=request))
|
|
122
|
+
assert connection.code is VoiceErrorCode.UNAVAILABLE and connection.retryable
|
|
123
|
+
|
|
124
|
+
|
|
125
|
+
# --- fakes, settings and factory ----------------------------------------------------------------
|
|
126
|
+
|
|
127
|
+
|
|
128
|
+
async def test_fake_synthesizer_returns_a_valid_wav_that_grows_with_the_text():
|
|
129
|
+
short = await FakeSynthesizer().synthesize("hola")
|
|
130
|
+
long = await FakeSynthesizer().synthesize("hola " * 20)
|
|
131
|
+
with wave.open(io.BytesIO(short.audio)) as wav:
|
|
132
|
+
assert wav.getnchannels() == 1 and wav.getframerate() == 8000 and wav.getnframes() == 8000 * 4 * 50 // 1000
|
|
133
|
+
assert len(long.audio) > len(short.audio) and short.mime == "audio/wav"
|
|
134
|
+
|
|
135
|
+
|
|
136
|
+
async def test_fake_transcriber_records_calls_and_silence():
|
|
137
|
+
transcriber = FakeTranscriber(text="dictado")
|
|
138
|
+
assert (await transcriber.transcribe(WEBM)).text == "dictado"
|
|
139
|
+
assert (await transcriber.transcribe(SILENCE)).text == ""
|
|
140
|
+
assert [a.mime for a, _ in transcriber.calls] == ["audio/webm", "audio/wav"]
|
|
141
|
+
|
|
142
|
+
|
|
143
|
+
def test_settings_require_providers_and_the_key_for_openai():
|
|
144
|
+
with pytest.raises(ValueError, match="stt_provider"):
|
|
145
|
+
VoiceSettings(_env_file=None, tts_provider="fake")
|
|
146
|
+
with pytest.raises(ValueError, match="OPENAI_API_KEY"):
|
|
147
|
+
settings(stt_provider="openai")
|
|
148
|
+
with pytest.raises(ValueError, match="OPENAI_API_KEY"):
|
|
149
|
+
settings(tts_provider="openai")
|
|
150
|
+
s = settings(stt_provider="openai", openai_api_key="sk-x", stt_language=None, tts_format="wav")
|
|
151
|
+
assert s.stt_provider is SttProvider.OPENAI and s.tts_provider is TtsProvider.FAKE and s.stt_language is None
|
|
152
|
+
assert s.stt_max_bytes == 10 * 1024 * 1024 and s.tts_max_chars == 4096 and s.tts_format == "wav"
|
|
153
|
+
|
|
154
|
+
|
|
155
|
+
def test_factory_builds_the_configured_providers_with_their_settings():
|
|
156
|
+
s = settings(stt_language="en", stt_prompt="p", stt_max_bytes=99, stt_timeout_seconds=1.5, tts_voice="v", tts_max_chars=10)
|
|
157
|
+
transcriber, synthesizer = VoiceFactory.transcriber(s), VoiceFactory.synthesizer(s)
|
|
158
|
+
assert isinstance(transcriber, FakeTranscriber) and (transcriber.language, transcriber.prompt, transcriber.max_bytes, transcriber.timeout_seconds) == ("en", "p", 99, 1.5)
|
|
159
|
+
assert isinstance(synthesizer, FakeSynthesizer) and (synthesizer.voice, synthesizer.max_chars) == ("v", 10)
|
|
160
|
+
|
|
161
|
+
o = settings(stt_provider="openai", tts_provider="openai", openai_api_key=SecretStr("sk-x"), stt_model="whisper-1", tts_format="opus")
|
|
162
|
+
assert isinstance(VoiceFactory.transcriber(o), OpenAITranscriber) and VoiceFactory.transcriber(o).model == "whisper-1"
|
|
163
|
+
synthesizer = VoiceFactory.synthesizer(o)
|
|
164
|
+
assert isinstance(synthesizer, OpenAISynthesizer) and synthesizer.mime == "audio/ogg" and "nova" in synthesizer.voices
|
|
@@ -0,0 +1,106 @@
|
|
|
1
|
+
"""Needs `fastapi-base` (build_app, ApiError) and `auth-verifier` (api/auth.py) copied next to this recipe."""
|
|
2
|
+
|
|
3
|
+
import httpx
|
|
4
|
+
import pytest
|
|
5
|
+
|
|
6
|
+
pytest.importorskip("api.base")
|
|
7
|
+
pytest.importorskip("api.auth")
|
|
8
|
+
|
|
9
|
+
from api.auth import CurrentUser, StaticTokenVerifier # noqa: E402
|
|
10
|
+
from api.base import build_app # noqa: E402
|
|
11
|
+
from api.voice.router import router # noqa: E402
|
|
12
|
+
from voice import VoiceError, VoiceErrorCode # noqa: E402
|
|
13
|
+
from voice.fake import FakeSynthesizer, FakeTranscriber # noqa: E402
|
|
14
|
+
|
|
15
|
+
AUTH = {"Authorization": "Bearer tok"}
|
|
16
|
+
|
|
17
|
+
|
|
18
|
+
def make_app(transcriber=None, synthesizer=None):
|
|
19
|
+
app = build_app(title="test")
|
|
20
|
+
app.state.token_verifier = StaticTokenVerifier({"tok": CurrentUser(id="u1", tenant_id="t1")})
|
|
21
|
+
app.state.transcriber = transcriber
|
|
22
|
+
app.state.synthesizer = synthesizer
|
|
23
|
+
app.include_router(router)
|
|
24
|
+
return app
|
|
25
|
+
|
|
26
|
+
|
|
27
|
+
def client(app) -> httpx.AsyncClient:
|
|
28
|
+
return httpx.AsyncClient(transport=httpx.ASGITransport(app=app), base_url="http://test")
|
|
29
|
+
|
|
30
|
+
|
|
31
|
+
def upload(content: bytes, mime: str = "audio/webm", **form):
|
|
32
|
+
return {"files": {"file": ("rec.webm", content, mime)}, "data": form}
|
|
33
|
+
|
|
34
|
+
|
|
35
|
+
async def test_transcribe_returns_text_and_stores_nothing():
|
|
36
|
+
transcriber = FakeTranscriber(text="hola mundo", language="es")
|
|
37
|
+
async with client(make_app(transcriber=transcriber)) as c:
|
|
38
|
+
response = await c.post("/voice/transcribe", headers=AUTH, **upload(b"\x1aE\xdf\xa3...", "audio/webm;codecs=opus", language="en"))
|
|
39
|
+
assert response.status_code == 200 and response.json() == {"text": "hola mundo", "language": "en"}
|
|
40
|
+
assert transcriber.calls[0][0].mime == "audio/webm" and transcriber.calls[0][1] == "en"
|
|
41
|
+
|
|
42
|
+
|
|
43
|
+
async def test_transcribe_requires_auth():
|
|
44
|
+
async with client(make_app(transcriber=FakeTranscriber())) as c:
|
|
45
|
+
assert (await c.post("/voice/transcribe", **upload(b"x"))).status_code == 401
|
|
46
|
+
|
|
47
|
+
|
|
48
|
+
@pytest.mark.parametrize(
|
|
49
|
+
"kwargs, status, code",
|
|
50
|
+
[
|
|
51
|
+
(upload(b"x", "audio/aac"), 415, "unsupported_audio_format"),
|
|
52
|
+
(upload(b"x" * 100, "audio/webm"), 413, "audio_too_large"),
|
|
53
|
+
(upload(b"", "audio/webm"), 422, "empty_audio"),
|
|
54
|
+
(upload(b"SILENCE", "audio/wav"), 422, "no_speech"),
|
|
55
|
+
(upload(b"x", "audio/webm", language="español"), 422, "invalid_language"),
|
|
56
|
+
],
|
|
57
|
+
)
|
|
58
|
+
async def test_transcribe_rejections_have_a_code(kwargs, status, code):
|
|
59
|
+
async with client(make_app(transcriber=FakeTranscriber(max_bytes=50))) as c:
|
|
60
|
+
response = await c.post("/voice/transcribe", headers=AUTH, **kwargs)
|
|
61
|
+
assert (response.status_code, response.json()["detail"]["code"]) == (status, code)
|
|
62
|
+
|
|
63
|
+
|
|
64
|
+
class Failing(FakeTranscriber):
|
|
65
|
+
def __init__(self, error: VoiceError) -> None:
|
|
66
|
+
super().__init__()
|
|
67
|
+
self.error = error
|
|
68
|
+
|
|
69
|
+
async def _transcribe(self, audio, language):
|
|
70
|
+
raise self.error
|
|
71
|
+
|
|
72
|
+
|
|
73
|
+
@pytest.mark.parametrize(
|
|
74
|
+
"error, status, code",
|
|
75
|
+
[
|
|
76
|
+
(VoiceError(VoiceErrorCode.RATE_LIMITED, retryable=True), 429, "provider_rate_limited"),
|
|
77
|
+
(VoiceError(VoiceErrorCode.TIMEOUT, retryable=True), 504, "provider_timeout"),
|
|
78
|
+
(VoiceError(VoiceErrorCode.UNAVAILABLE, retryable=True), 502, "provider_unavailable"),
|
|
79
|
+
(VoiceError(VoiceErrorCode.UNAVAILABLE, retryable=False), 503, "provider_unavailable"),
|
|
80
|
+
(VoiceError(VoiceErrorCode.INVALID_AUDIO, retryable=False), 422, "invalid_audio"),
|
|
81
|
+
],
|
|
82
|
+
)
|
|
83
|
+
async def test_provider_failures_map_to_http(error, status, code):
|
|
84
|
+
async with client(make_app(transcriber=Failing(error))) as c:
|
|
85
|
+
response = await c.post("/voice/transcribe", headers=AUTH, **upload(b"x"))
|
|
86
|
+
assert (response.status_code, response.json()["detail"]["code"]) == (status, code)
|
|
87
|
+
assert "x-request-id" in response.headers
|
|
88
|
+
if status == 429:
|
|
89
|
+
assert response.headers["retry-after"] == "5"
|
|
90
|
+
|
|
91
|
+
|
|
92
|
+
async def test_synthesize_returns_audio_with_no_store():
|
|
93
|
+
async with client(make_app(synthesizer=FakeSynthesizer())) as c:
|
|
94
|
+
ok = await c.post("/voice/synthesize", headers=AUTH, json={"text": "Hola", "voice": "any"})
|
|
95
|
+
empty = await c.post("/voice/synthesize", headers=AUTH, json={"text": " "})
|
|
96
|
+
unauth = await c.post("/voice/synthesize", json={"text": "Hola"})
|
|
97
|
+
assert ok.status_code == 200 and ok.headers["content-type"] == "audio/wav" and ok.headers["cache-control"] == "no-store"
|
|
98
|
+
assert ok.content[:4] == b"RIFF"
|
|
99
|
+
assert (empty.status_code, empty.json()["detail"]["code"]) == (422, "empty_text") and unauth.status_code == 401
|
|
100
|
+
|
|
101
|
+
|
|
102
|
+
async def test_app_without_voice_answers_503():
|
|
103
|
+
async with client(make_app()) as c:
|
|
104
|
+
t = await c.post("/voice/transcribe", headers=AUTH, **upload(b"x"))
|
|
105
|
+
s = await c.post("/voice/synthesize", headers=AUTH, json={"text": "Hola"})
|
|
106
|
+
assert t.status_code == 503 and s.status_code == 503 and t.json()["detail"]["code"] == "voice_unavailable"
|
|
@@ -0,0 +1,80 @@
|
|
|
1
|
+
"""Behaviour every `Transcriber` / `Synthesizer` must have. A new provider = a subclass with the fixture.
|
|
2
|
+
|
|
3
|
+
class TestDeepgramTranscriber(TranscriberConformance):
|
|
4
|
+
@pytest.fixture
|
|
5
|
+
def transcriber(self): return DeepgramTranscriber(...) # must answer "hola mundo" for WEBM and "" for SILENCE
|
|
6
|
+
"""
|
|
7
|
+
|
|
8
|
+
import asyncio
|
|
9
|
+
|
|
10
|
+
import pytest
|
|
11
|
+
|
|
12
|
+
from voice import AudioInput, Speech, Transcription, VoiceError, VoiceErrorCode
|
|
13
|
+
|
|
14
|
+
WEBM = AudioInput(content=b"\x1aE\xdf\xa3fake-webm", mime="audio/webm")
|
|
15
|
+
SILENCE = AudioInput(content=b"SILENCE", mime="audio/wav")
|
|
16
|
+
|
|
17
|
+
|
|
18
|
+
class TranscriberConformance:
|
|
19
|
+
async def test_returns_a_transcription_with_engine_and_timing(self, transcriber):
|
|
20
|
+
result = await transcriber.transcribe(WEBM)
|
|
21
|
+
assert isinstance(result, Transcription) and result.text == "hola mundo" and result.engine
|
|
22
|
+
assert result.duration_ms >= 0 and result.language == transcriber.language
|
|
23
|
+
|
|
24
|
+
async def test_silence_is_empty_text_not_an_error(self, transcriber):
|
|
25
|
+
assert (await transcriber.transcribe(SILENCE)).text == ""
|
|
26
|
+
|
|
27
|
+
async def test_language_of_the_call_wins(self, transcriber):
|
|
28
|
+
assert (await transcriber.transcribe(WEBM, language="en")).language == "en"
|
|
29
|
+
|
|
30
|
+
@pytest.mark.parametrize(
|
|
31
|
+
"audio, kind",
|
|
32
|
+
[
|
|
33
|
+
(AudioInput(content=b"", mime="audio/webm"), "EmptyAudio"),
|
|
34
|
+
(AudioInput(content=b"x", mime="audio/aac"), "UnsupportedFormat"),
|
|
35
|
+
(AudioInput(content=b"x", mime="text/plain"), "UnsupportedFormat"),
|
|
36
|
+
],
|
|
37
|
+
)
|
|
38
|
+
async def test_invalid_audio_is_rejected_before_the_provider(self, transcriber, audio, kind):
|
|
39
|
+
with pytest.raises(VoiceError) as info:
|
|
40
|
+
await transcriber.transcribe(audio)
|
|
41
|
+
assert info.value.code is VoiceErrorCode.INVALID_AUDIO and info.value.kind == kind and not info.value.retryable
|
|
42
|
+
|
|
43
|
+
async def test_size_limit(self, transcriber):
|
|
44
|
+
transcriber.max_bytes = 4
|
|
45
|
+
with pytest.raises(VoiceError, match="AudioTooLarge"):
|
|
46
|
+
await transcriber.transcribe(WEBM)
|
|
47
|
+
|
|
48
|
+
async def test_timeout_becomes_a_retryable_voice_error(self, transcriber, monkeypatch):
|
|
49
|
+
async def slow(*args, **kwargs):
|
|
50
|
+
await asyncio.sleep(1)
|
|
51
|
+
return "late"
|
|
52
|
+
|
|
53
|
+
monkeypatch.setattr(transcriber, "_transcribe", slow)
|
|
54
|
+
transcriber.timeout_seconds = 0.01
|
|
55
|
+
with pytest.raises(VoiceError) as info:
|
|
56
|
+
await transcriber.transcribe(WEBM)
|
|
57
|
+
assert info.value.code is VoiceErrorCode.TIMEOUT and info.value.retryable
|
|
58
|
+
|
|
59
|
+
|
|
60
|
+
class SynthesizerConformance:
|
|
61
|
+
async def test_returns_playable_bytes_with_mime(self, synthesizer):
|
|
62
|
+
speech = await synthesizer.synthesize("Hola, ¿qué tal?")
|
|
63
|
+
assert isinstance(speech, Speech) and len(speech.audio) > 0 and speech.mime.startswith("audio/")
|
|
64
|
+
assert speech.engine and speech.duration_ms >= 0
|
|
65
|
+
|
|
66
|
+
async def test_empty_and_too_long_text(self, synthesizer):
|
|
67
|
+
with pytest.raises(VoiceError) as empty:
|
|
68
|
+
await synthesizer.synthesize(" ")
|
|
69
|
+
assert empty.value.code is VoiceErrorCode.EMPTY_TEXT
|
|
70
|
+
synthesizer.max_chars = 5
|
|
71
|
+
with pytest.raises(VoiceError) as long:
|
|
72
|
+
await synthesizer.synthesize("demasiado largo")
|
|
73
|
+
assert long.value.code is VoiceErrorCode.TEXT_TOO_LONG and not long.value.retryable
|
|
74
|
+
|
|
75
|
+
async def test_unknown_voice_is_rejected_when_there_is_an_allowlist(self, synthesizer):
|
|
76
|
+
if not synthesizer.voices:
|
|
77
|
+
pytest.skip("this provider accepts any voice")
|
|
78
|
+
with pytest.raises(VoiceError) as info:
|
|
79
|
+
await synthesizer.synthesize("hola", voice="not-a-voice")
|
|
80
|
+
assert info.value.code is VoiceErrorCode.INVALID_VOICE
|
|
@@ -0,0 +1,29 @@
|
|
|
1
|
+
"""Speech to text and text to speech with the provider chosen in settings.
|
|
2
|
+
|
|
3
|
+
transcriber = VoiceFactory.transcriber(settings) # startup, once; app.state.transcriber
|
|
4
|
+
result = await transcriber.transcribe(AudioInput(content=data, mime="audio/webm"))
|
|
5
|
+
if not result.text: ... # no speech
|
|
6
|
+
speech = await synthesizer.synthesize("Hola") # Speech(audio=bytes, mime="audio/mpeg")
|
|
7
|
+
|
|
8
|
+
Every failure is a `VoiceError(code, retryable)`; `api/voice/router.py` maps the codes to HTTP.
|
|
9
|
+
"""
|
|
10
|
+
|
|
11
|
+
from .base import Synthesizer, Transcriber
|
|
12
|
+
from .factory import VoiceFactory
|
|
13
|
+
from .models import AUDIO_FORMATS, AudioInput, Speech, SttProvider, Transcription, TtsProvider, VoiceError, VoiceErrorCode
|
|
14
|
+
from .settings import VoiceSettings
|
|
15
|
+
|
|
16
|
+
__all__ = [
|
|
17
|
+
"AUDIO_FORMATS",
|
|
18
|
+
"AudioInput",
|
|
19
|
+
"Speech",
|
|
20
|
+
"SttProvider",
|
|
21
|
+
"Synthesizer",
|
|
22
|
+
"Transcriber",
|
|
23
|
+
"Transcription",
|
|
24
|
+
"TtsProvider",
|
|
25
|
+
"VoiceError",
|
|
26
|
+
"VoiceErrorCode",
|
|
27
|
+
"VoiceFactory",
|
|
28
|
+
"VoiceSettings",
|
|
29
|
+
]
|
|
@@ -0,0 +1,116 @@
|
|
|
1
|
+
"""`Transcriber` (speech -> text) and `Synthesizer` (text -> speech): the contracts.
|
|
2
|
+
|
|
3
|
+
Both are template methods: the public call validates the input, measures, bounds the time and
|
|
4
|
+
translates every failure into `VoiceError`; the subclass only talks to its provider in
|
|
5
|
+
`_transcribe` / `_synthesize`. Nothing here retries: `VoiceError.retryable` tells the caller.
|
|
6
|
+
"""
|
|
7
|
+
|
|
8
|
+
import asyncio
|
|
9
|
+
import logging
|
|
10
|
+
import time
|
|
11
|
+
from abc import ABC, abstractmethod
|
|
12
|
+
from collections.abc import Iterable
|
|
13
|
+
|
|
14
|
+
from .models import AudioInput, Speech, Transcription, VoiceError, VoiceErrorCode
|
|
15
|
+
|
|
16
|
+
logger = logging.getLogger(__name__)
|
|
17
|
+
|
|
18
|
+
|
|
19
|
+
class Transcriber(ABC):
|
|
20
|
+
engine: str = "transcriber"
|
|
21
|
+
|
|
22
|
+
def __init__(
|
|
23
|
+
self,
|
|
24
|
+
*,
|
|
25
|
+
language: str | None = None,
|
|
26
|
+
prompt: str | None = None,
|
|
27
|
+
max_bytes: int = 10 * 1024 * 1024,
|
|
28
|
+
timeout_seconds: float = 60.0,
|
|
29
|
+
) -> None:
|
|
30
|
+
self.language = language
|
|
31
|
+
self.prompt = prompt
|
|
32
|
+
self.max_bytes = max_bytes
|
|
33
|
+
self.timeout_seconds = timeout_seconds
|
|
34
|
+
|
|
35
|
+
async def transcribe(self, audio: AudioInput, *, language: str | None = None) -> Transcription:
|
|
36
|
+
"""`language=None` -> the one configured. Empty text = no speech."""
|
|
37
|
+
if not audio.content:
|
|
38
|
+
raise VoiceError(VoiceErrorCode.INVALID_AUDIO, retryable=False, kind="EmptyAudio")
|
|
39
|
+
if audio.extension is None:
|
|
40
|
+
raise VoiceError(VoiceErrorCode.INVALID_AUDIO, retryable=False, kind="UnsupportedFormat")
|
|
41
|
+
if len(audio.content) > self.max_bytes:
|
|
42
|
+
raise VoiceError(VoiceErrorCode.INVALID_AUDIO, retryable=False, kind="AudioTooLarge")
|
|
43
|
+
lang = language or self.language
|
|
44
|
+
started = time.monotonic()
|
|
45
|
+
text = await _bounded(self._transcribe(audio, lang), self.timeout_seconds, self.engine, self._translate)
|
|
46
|
+
if not isinstance(text, str):
|
|
47
|
+
raise VoiceError(VoiceErrorCode.UNAVAILABLE, retryable=False, kind="InvalidProviderOutput")
|
|
48
|
+
return Transcription(
|
|
49
|
+
text=text.strip(), language=lang, engine=self.engine, duration_ms=int((time.monotonic() - started) * 1000)
|
|
50
|
+
)
|
|
51
|
+
|
|
52
|
+
@abstractmethod
|
|
53
|
+
async def _transcribe(self, audio: AudioInput, language: str | None) -> str: ...
|
|
54
|
+
|
|
55
|
+
def _translate(self, error: Exception) -> VoiceError:
|
|
56
|
+
"""Map the provider's exceptions to `VoiceError`. Override per provider."""
|
|
57
|
+
return VoiceError(VoiceErrorCode.UNAVAILABLE, retryable=True, kind=type(error).__name__)
|
|
58
|
+
|
|
59
|
+
async def close(self) -> None:
|
|
60
|
+
return None
|
|
61
|
+
|
|
62
|
+
|
|
63
|
+
class Synthesizer(ABC):
|
|
64
|
+
engine: str = "synthesizer"
|
|
65
|
+
mime: str = "audio/mpeg"
|
|
66
|
+
|
|
67
|
+
def __init__(
|
|
68
|
+
self,
|
|
69
|
+
*,
|
|
70
|
+
voice: str,
|
|
71
|
+
voices: Iterable[str] = (),
|
|
72
|
+
max_chars: int = 4096,
|
|
73
|
+
timeout_seconds: float = 60.0,
|
|
74
|
+
) -> None:
|
|
75
|
+
self.voice = voice
|
|
76
|
+
self.voices = frozenset(voices) # empty = any voice is accepted
|
|
77
|
+
self.max_chars = max_chars
|
|
78
|
+
self.timeout_seconds = timeout_seconds
|
|
79
|
+
|
|
80
|
+
async def synthesize(self, text: str, *, voice: str | None = None) -> Speech:
|
|
81
|
+
text = text.strip()
|
|
82
|
+
if not text:
|
|
83
|
+
raise VoiceError(VoiceErrorCode.EMPTY_TEXT, retryable=False)
|
|
84
|
+
if len(text) > self.max_chars:
|
|
85
|
+
raise VoiceError(VoiceErrorCode.TEXT_TOO_LONG, retryable=False)
|
|
86
|
+
chosen = voice or self.voice
|
|
87
|
+
if self.voices and chosen not in self.voices:
|
|
88
|
+
raise VoiceError(VoiceErrorCode.INVALID_VOICE, retryable=False)
|
|
89
|
+
started = time.monotonic()
|
|
90
|
+
audio = await _bounded(self._synthesize(text, chosen), self.timeout_seconds, self.engine, self._translate)
|
|
91
|
+
if not isinstance(audio, bytes | bytearray) or not audio:
|
|
92
|
+
raise VoiceError(VoiceErrorCode.UNAVAILABLE, retryable=False, kind="InvalidProviderOutput")
|
|
93
|
+
return Speech(audio=bytes(audio), mime=self.mime, engine=self.engine, duration_ms=int((time.monotonic() - started) * 1000))
|
|
94
|
+
|
|
95
|
+
@abstractmethod
|
|
96
|
+
async def _synthesize(self, text: str, voice: str) -> bytes: ...
|
|
97
|
+
|
|
98
|
+
def _translate(self, error: Exception) -> VoiceError:
|
|
99
|
+
return VoiceError(VoiceErrorCode.UNAVAILABLE, retryable=True, kind=type(error).__name__)
|
|
100
|
+
|
|
101
|
+
async def close(self) -> None:
|
|
102
|
+
return None
|
|
103
|
+
|
|
104
|
+
|
|
105
|
+
async def _bounded(awaitable, timeout: float, engine: str, translate):
|
|
106
|
+
try:
|
|
107
|
+
return await asyncio.wait_for(awaitable, timeout=timeout)
|
|
108
|
+
except VoiceError:
|
|
109
|
+
raise
|
|
110
|
+
except TimeoutError:
|
|
111
|
+
logger.warning("%s timed out after %.0fs", engine, timeout)
|
|
112
|
+
raise VoiceError(VoiceErrorCode.TIMEOUT, retryable=True, kind="TimeoutError") from None
|
|
113
|
+
except Exception as exc: # the message may hold the key or the URL: class name only
|
|
114
|
+
error = translate(exc)
|
|
115
|
+
logger.warning("%s failed: %s -> %s", engine, type(exc).__name__, error.code.value)
|
|
116
|
+
raise error from None
|
|
@@ -0,0 +1,48 @@
|
|
|
1
|
+
"""The one place that builds the transcriber and the synthesizer the settings name."""
|
|
2
|
+
|
|
3
|
+
from typing import Any
|
|
4
|
+
|
|
5
|
+
from .base import Synthesizer, Transcriber
|
|
6
|
+
from .models import SttProvider, TtsProvider
|
|
7
|
+
from .settings import VoiceSettings
|
|
8
|
+
|
|
9
|
+
|
|
10
|
+
class VoiceFactory:
|
|
11
|
+
@staticmethod
|
|
12
|
+
def transcriber(settings: VoiceSettings, *, http_client: Any | None = None) -> Transcriber:
|
|
13
|
+
"""Call it once at startup: `app.state.transcriber = VoiceFactory.transcriber(settings)`.
|
|
14
|
+
`http_client` is the test seam (httpx.MockTransport). Unknown provider -> ValueError."""
|
|
15
|
+
common = {
|
|
16
|
+
"language": settings.stt_language,
|
|
17
|
+
"prompt": settings.stt_prompt,
|
|
18
|
+
"max_bytes": settings.stt_max_bytes,
|
|
19
|
+
"timeout_seconds": settings.stt_timeout_seconds,
|
|
20
|
+
}
|
|
21
|
+
match settings.stt_provider:
|
|
22
|
+
case SttProvider.FAKE:
|
|
23
|
+
from .fake import FakeTranscriber
|
|
24
|
+
|
|
25
|
+
return FakeTranscriber(**common)
|
|
26
|
+
case SttProvider.OPENAI:
|
|
27
|
+
from .openai_provider import OpenAITranscriber, openai_client
|
|
28
|
+
|
|
29
|
+
assert settings.openai_api_key is not None # the settings validator guarantees it
|
|
30
|
+
client = openai_client(settings.openai_api_key, base_url=settings.openai_base_url, http_client=http_client)
|
|
31
|
+
return OpenAITranscriber(client, model=settings.stt_model, no_speech_threshold=settings.stt_no_speech_threshold, **common)
|
|
32
|
+
raise ValueError(f"Unknown STT provider: {settings.stt_provider!r}")
|
|
33
|
+
|
|
34
|
+
@staticmethod
|
|
35
|
+
def synthesizer(settings: VoiceSettings, *, http_client: Any | None = None) -> Synthesizer:
|
|
36
|
+
common = {"voice": settings.tts_voice, "max_chars": settings.tts_max_chars, "timeout_seconds": settings.tts_timeout_seconds}
|
|
37
|
+
match settings.tts_provider:
|
|
38
|
+
case TtsProvider.FAKE:
|
|
39
|
+
from .fake import FakeSynthesizer
|
|
40
|
+
|
|
41
|
+
return FakeSynthesizer(**common)
|
|
42
|
+
case TtsProvider.OPENAI:
|
|
43
|
+
from .openai_provider import OpenAISynthesizer, openai_client
|
|
44
|
+
|
|
45
|
+
assert settings.openai_api_key is not None
|
|
46
|
+
client = openai_client(settings.openai_api_key, base_url=settings.openai_base_url, http_client=http_client)
|
|
47
|
+
return OpenAISynthesizer(client, model=settings.tts_model, audio_format=settings.tts_format, **common)
|
|
48
|
+
raise ValueError(f"Unknown TTS provider: {settings.tts_provider!r}")
|
|
@@ -0,0 +1,51 @@
|
|
|
1
|
+
"""Test doubles and local provider (`STT_PROVIDER=fake`): no network, deterministic.
|
|
2
|
+
|
|
3
|
+
`FakeTranscriber` returns a fixed text, or "" when the audio is the literal `b"SILENCE"`.
|
|
4
|
+
`FakeSynthesizer` returns a valid WAV of silence whose length grows with the text, so a client
|
|
5
|
+
can actually play it.
|
|
6
|
+
"""
|
|
7
|
+
|
|
8
|
+
import io
|
|
9
|
+
import struct
|
|
10
|
+
import wave
|
|
11
|
+
|
|
12
|
+
from .base import Synthesizer, Transcriber
|
|
13
|
+
from .models import AudioInput
|
|
14
|
+
|
|
15
|
+
|
|
16
|
+
class FakeTranscriber(Transcriber):
|
|
17
|
+
engine = "fake"
|
|
18
|
+
|
|
19
|
+
def __init__(self, text: str = "test transcription", **kwargs) -> None:
|
|
20
|
+
super().__init__(**kwargs)
|
|
21
|
+
self.text = text
|
|
22
|
+
self.calls: list[tuple[AudioInput, str | None]] = []
|
|
23
|
+
|
|
24
|
+
async def _transcribe(self, audio: AudioInput, language: str | None) -> str:
|
|
25
|
+
self.calls.append((audio, language))
|
|
26
|
+
return "" if audio.content == b"SILENCE" else self.text
|
|
27
|
+
|
|
28
|
+
|
|
29
|
+
class FakeSynthesizer(Synthesizer):
|
|
30
|
+
engine = "fake"
|
|
31
|
+
mime = "audio/wav"
|
|
32
|
+
|
|
33
|
+
def __init__(self, *, voice: str = "fake", ms_per_char: int = 50, **kwargs) -> None:
|
|
34
|
+
super().__init__(voice=voice, **kwargs)
|
|
35
|
+
self.ms_per_char = ms_per_char
|
|
36
|
+
self.calls: list[tuple[str, str]] = []
|
|
37
|
+
|
|
38
|
+
async def _synthesize(self, text: str, voice: str) -> bytes:
|
|
39
|
+
self.calls.append((text, voice))
|
|
40
|
+
return silence_wav(milliseconds=len(text) * self.ms_per_char)
|
|
41
|
+
|
|
42
|
+
|
|
43
|
+
def silence_wav(*, milliseconds: int, sample_rate: int = 8000) -> bytes:
|
|
44
|
+
frames = max(1, sample_rate * milliseconds // 1000)
|
|
45
|
+
buffer = io.BytesIO()
|
|
46
|
+
with wave.open(buffer, "wb") as wav:
|
|
47
|
+
wav.setnchannels(1)
|
|
48
|
+
wav.setsampwidth(2)
|
|
49
|
+
wav.setframerate(sample_rate)
|
|
50
|
+
wav.writeframes(struct.pack(f"<{frames}h", *([0] * frames)))
|
|
51
|
+
return buffer.getvalue()
|