Class: AssemblyaiClient

Inherits:
Object
  • Object
show all
Includes:
Singleton
Defined in:
app/services/assemblyai_client.rb

Constant Summary collapse

BASE_URL = 
'https://api.assemblyai.com/v2'
LLM_GATEWAY_URL = 
'https://llm-gateway.assemblyai.com/v1'

Instance Attribute Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(options = {}) ⇒ AssemblyaiClient

Returns a new instance of AssemblyaiClient.



13
14
15
16
17
18
19
20
 
# File 'app/services/assemblyai_client.rb', line 13

def initialize(options = {})
  @logger = options[:logger] || Rails.logger
  @api_key = Heatwave::Configuration.fetch(:assemblyai, :api_key)

  @client = HTTPClient.new
  @client.default_header['Authorization'] = @api_key
  @client.default_header['Content-Type'] = 'application/json'
end

Instance Attribute Details

#api_key ⇒ Object (readonly)

Returns the value of attribute api_key.



11
12
13
 
# File 'app/services/assemblyai_client.rb', line 11

def api_key
  @api_key
end

#client ⇒ Object (readonly)

Returns the value of attribute client.



11
12
13
 
# File 'app/services/assemblyai_client.rb', line 11

def client
  @client
end

#logger ⇒ Object (readonly)

Returns the value of attribute logger.



11
12
13
 
# File 'app/services/assemblyai_client.rb', line 11

def logger
  @logger
end

Instance Method Details

#export_paragraphs(transcript_id) ⇒ Object 



171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
 
# File 'app/services/assemblyai_client.rb', line 171

def export_paragraphs(transcript_id)
  logger.info "[assemblyai:export_paragraphs] Exporting paragraphs for transcript: #{transcript_id}"
  export_url = "#{BASE_URL}/transcript/#{transcript_id}/paragraphs"
  begin
    response = @client.get(export_url)
    if response.status == 200
      result = JSON.parse(response.body)
      logger.info "[assemblyai:export_paragraphs] Success: #{result['paragraphs']&.length || 0} paragraphs"
      result
    else
      logger.error "[assemblyai:export_paragraphs] Failed with status #{response.status}: #{response.body}"
      raise "AssemblyAI paragraphs export failed: #{response.status} - #{response.body}"
    end
  rescue StandardError => e
    logger.error "[assemblyai:export_paragraphs] Exception: #{e.message}"
    raise "AssemblyAI paragraphs export failed: #{e.message}"
  end
end

#export_sentences(transcript_id) ⇒ Object

Export transcript as sentences



191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
 
# File 'app/services/assemblyai_client.rb', line 191

def export_sentences(transcript_id)
  logger.info "[assemblyai:export_sentences] Exporting sentences for transcript: #{transcript_id}"

  export_url = "#{BASE_URL}/transcript/#{transcript_id}/sentences"

  begin
    response = @client.get(export_url)

    if response.status == 200
      result = JSON.parse(response.body)
      logger.info "[assemblyai:export_sentences] Success: #{result['sentences']&.length || 0} sentences"
      result
    else
      logger.error "[assemblyai:export_sentences] Failed with status #{response.status}: #{response.body}"
      raise "AssemblyAI sentences export failed: #{response.status} - #{response.body}"
    end
  rescue StandardError => e
    logger.error "[assemblyai:export_sentences] Exception: #{e.message}"
    raise "AssemblyAI sentences export failed: #{e.message}"
  end
end

#export_vtt(transcript_id, chars_per_caption = 32) ⇒ Object

Export transcript as VTT captions



214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
 
# File 'app/services/assemblyai_client.rb', line 214

def export_vtt(transcript_id, chars_per_caption = 32)
  logger.info "[assemblyai:export_vtt] Exporting VTT for transcript: #{transcript_id}"

  export_url = "#{BASE_URL}/transcript/#{transcript_id}/vtt"

  # Add query parameters for customization
  params = { chars_per_caption: chars_per_caption }
  query_string = params.map { |k, v| "#{k}=#{v}" }.join('&')
  full_url = query_string.present? ? "#{export_url}?#{query_string}" : export_url

  begin
    response = @client.get(full_url)

    if response.status == 200
      logger.info "[assemblyai:export_vtt] Success: #{response.body.length} characters"
      response.body
    else
      logger.error "[assemblyai:export_vtt] Failed with status #{response.status}: #{response.body}"
      raise "AssemblyAI VTT export failed: #{response.status} - #{response.body}"
    end
  rescue StandardError => e
    logger.error "[assemblyai:export_vtt] Exception: #{e.message}"
    raise "AssemblyAI VTT export failed: #{e.message}"
  end
end

#get_transcription(transcript_id) ⇒ Object

Get transcription result



89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
 
# File 'app/services/assemblyai_client.rb', line 89

def get_transcription(transcript_id)
  logger.info "[assemblyai:get_transcription] Getting result for: #{transcript_id}"

  transcript_url = "#{BASE_URL}/transcript/#{transcript_id}"

  begin
    response = @client.get(transcript_url)

    if response.status == 200
      result = JSON.parse(response.body)
      logger.info "[assemblyai:get_transcription] Success: #{result['status']}"
      result
    else
      logger.error "[assemblyai:get_transcription] Failed with status #{response.status}: #{response.body}"
      raise "AssemblyAI get transcription failed: #{response.status} - #{response.body}"
    end
  rescue StandardError => e
    logger.error "[assemblyai:get_transcription] Exception: #{e.message}"
    raise "AssemblyAI get transcription failed: #{e.message}"
  end
end

#identify_speakers(transcript_id:, known_values:, speaker_type: 'name') ⇒ Hash

Speaker Identification via AssemblyAI Speech Understanding API (Method 2 - post-transcription).
Takes a completed transcript ID and returns utterances with name/role labels instead of A/B.
Use this to identify speakers on an already-completed transcript without re-transcribing.

Parameters:

  • transcript_id (String) —

    Completed AssemblyAI transcript ID

  • known_values (Array<String>) —

    Speaker names or roles (max 35 chars each)

  • speaker_type (String) (defaults to: 'name') —

    'name' or 'role' (default: 'name')

Returns:

  • (Hash) —

    Response with updated utterances keyed by speaker name/role



397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
 
# File 'app/services/assemblyai_client.rb', line 397

def identify_speakers(transcript_id:, known_values:, speaker_type: 'name')
  logger.info "[assemblyai:identify_speakers] Running speaker identification for transcript #{transcript_id}"
  logger.info "[assemblyai:identify_speakers] known_values: #{known_values.inspect}"

  request_body = {
    transcript_id: transcript_id,
    speech_understanding: {
      request: {
        speaker_identification: {
          speaker_type: speaker_type,
          known_values: known_values.map { |v| v.to_s.gsub(/[[:space:]]/, ' ').squeeze(' ').strip.slice(0, 35) }.reject(&:blank?)
        }
      }
    }
  }

  begin
    response = @client.post("#{LLM_GATEWAY_URL}/understanding", request_body.to_json, nil, nil, 120)

    if response.status == 200
      result = JSON.parse(response.body)
      utterance_count = result['utterances']&.length || 0
      logger.info "[assemblyai:identify_speakers] Success - #{utterance_count} utterances returned"
      result
    else
      error_body = response.body.to_s.truncate(500)
      logger.error "[assemblyai:identify_speakers] Failed with status #{response.status}: #{error_body}"
      raise "Speaker identification failed: #{response.status} - #{error_body}"
    end
  rescue StandardError => e
    logger.error "[assemblyai:identify_speakers] Exception: #{e.message}"
    raise "Speaker identification failed: #{e.message}"
  end
end

#llm_gateway_chat(messages:, model: 'anthropic/claude-3-5-sonnet', max_tokens: 2000, temperature: 0.3, tools: nil, tool_choice: nil) ⇒ Object

LLM Gateway chat completion with transcript context
Uses the LLM Gateway to analyze transcripts with any supported model

Parameters:

  • messages (Array) —

    Array of message hashes with :role and :content

  • model (String) (defaults to: 'anthropic/claude-3-5-sonnet') —

    Model identifier (e.g., 'claude-sonnet-4-6-20260217')

  • max_tokens (Integer) (defaults to: 2000) —

    Maximum tokens in response

  • temperature (Float) (defaults to: 0.3) —

    Sampling temperature (0-2)

  • tools (Array) (defaults to: nil) —

    Optional array of tool definitions for structured output

  • tool_choice (String, Hash) (defaults to: nil) —

    Optional tool choice ('auto', 'none', or specific tool)



253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
 
# File 'app/services/assemblyai_client.rb', line 253

def llm_gateway_chat(messages:, model: 'anthropic/claude-3-5-sonnet', max_tokens: 2000, temperature: 0.3, tools: nil, tool_choice: nil)
  logger.info "[assemblyai:llm_gateway] Running chat completion with model: #{model}"

  request_body = {
    model: model,
    messages: messages,
    max_tokens: max_tokens,
    temperature: temperature
  }

  # Add tools for structured output if provided
  if tools.present?
    request_body[:tools] = tools
    request_body[:tool_choice] = tool_choice if tool_choice.present?
  end

  begin
    response = @client.post("#{LLM_GATEWAY_URL}/chat/completions", request_body.to_json)

    if response.status == 200
      result = JSON.parse(response.body)
      logger.info '[assemblyai:llm_gateway] Success'
      result
    else
      logger.error "[assemblyai:llm_gateway] Failed with status #{response.status}: #{response.body}"
      raise "LLM Gateway failed: #{response.status} - #{response.body}"
    end
  rescue StandardError => e
    logger.error "[assemblyai:llm_gateway] Exception: #{e.message}"
    raise "LLM Gateway failed: #{e.message}"
  end
end

#llm_gateway_structured(messages:, schema:, tool_name: 'structured_output', model: LlmDefaults::DEFAULT_SONNET_MODEL, max_tokens: 2000, temperature: 0.3) ⇒ Hash

LLM Gateway with structured output via tool calling
Forces the model to return data matching the provided JSON Schema

Parameters:

  • messages (Array) —

    Array of message hashes

  • schema (Hash) —

    JSON Schema defining the expected output structure

  • tool_name (String) (defaults to: 'structured_output') —

    Name for the tool (default: 'structured_output')

  • model (String) (defaults to: LlmDefaults::DEFAULT_SONNET_MODEL) —

    Model identifier

  • max_tokens (Integer) (defaults to: 2000) —

    Maximum tokens

  • temperature (Float) (defaults to: 0.3) —

    Sampling temperature

Returns:

  • (Hash) —

    Parsed structured response matching the schema



295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
 
# File 'app/services/assemblyai_client.rb', line 295

def llm_gateway_structured(messages:, schema:, tool_name: 'structured_output', model: LlmDefaults::DEFAULT_SONNET_MODEL, max_tokens: 2000, temperature: 0.3)
  logger.info "[assemblyai:llm_gateway_structured] Running structured output with model: #{model}"

  # Define a tool with the schema as parameters
  tools = [
    {
      type: 'function',
      function: {
        name: tool_name,
        description: 'Return the structured output',
        parameters: schema
      }
    }
  ]

  # Force the model to call this specific tool
  tool_choice = {
    type: 'function',
    function: { name: tool_name }
  }

  result = llm_gateway_chat(
    messages: messages,
    model: model,
    max_tokens: max_tokens,
    temperature: temperature,
    tools: tools,
    tool_choice: tool_choice
  )

  # Log the response structure for debugging
  choice = result.dig('choices', 0)
  logger.debug "[assemblyai:llm_gateway_structured] Response choice keys: #{choice&.keys&.join(', ')}"
  logger.debug "[assemblyai:llm_gateway_structured] finish_reason: #{choice&.dig('finish_reason')}"
  logger.debug "[assemblyai:llm_gateway_structured] message keys: #{choice&.dig('message')&.keys&.join(', ')}"

  # Extract the structured arguments from the tool call
  # Claude via AssemblyAI may use 'tool_use' content blocks instead of 'tool_calls'
  # Check various locations where tool calls may appear
  tool_calls = choice&.dig('tool_calls') ||
               choice&.dig('message', 'tool_calls') ||
               choice&.dig('message', 'tool_use')

  # Claude's Anthropic format uses content blocks with type: tool_use
  if tool_calls.blank?
    content_blocks = choice&.dig('message', 'content')
    if content_blocks.is_a?(Array)
      tool_use_block = content_blocks.find { |block| block['type'] == 'tool_use' }
      if tool_use_block.present?
        logger.info '[assemblyai:llm_gateway_structured] Found tool_use in content blocks'
        input = tool_use_block['input']
        if input.is_a?(Hash)
          logger.info '[assemblyai:llm_gateway_structured] Successfully extracted structured output from tool_use block'
          return input
        elsif input.is_a?(String)
          parsed = JSON.parse(input)
          logger.info '[assemblyai:llm_gateway_structured] Successfully parsed tool_use input string'
          return parsed
        end
      end
    end
  end

  if tool_calls.present? && tool_calls.first.present?
    tool_call = tool_calls.first
    # Check both 'function.arguments' (OpenAI format) and 'input' (Anthropic format)
    arguments = tool_call.dig('function', 'arguments') || tool_call['input']

    if arguments.present?
      if arguments.is_a?(Hash)
        # Already parsed - return directly
        logger.info '[assemblyai:llm_gateway_structured] Successfully extracted structured output from tool_calls (Hash)'
        return arguments
      elsif arguments.is_a?(String)
        # Parse JSON string
        parsed = JSON.parse(arguments)
        logger.info '[assemblyai:llm_gateway_structured] Successfully extracted structured output from tool_calls (parsed JSON)'
        return parsed
      end
    end
  end

  # Fallback: try to parse content as JSON (some models may return directly)
  content = choice&.dig('message', 'content')
  if content.is_a?(String) && content.present?
    logger.warn "[assemblyai:llm_gateway_structured] No tool_calls found, attempting to parse string content (length: #{content.length})"
    cleaned = content.gsub(/```json\n?/, '').gsub(/```\n?/, '').strip
    return JSON.parse(cleaned)
  end

  # Log the full choice structure for debugging
  logger.error "[assemblyai:llm_gateway_structured] No valid response found. Full choice: #{choice.to_json.truncate(1000)}"
  raise 'LLM Gateway structured output: No valid response found'
end

#poll_transcription(transcript_id, max_wait_time = 600, progress_callback = nil) ⇒ Object

Poll for transcription completion



112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
 
# File 'app/services/assemblyai_client.rb', line 112

def poll_transcription(transcript_id, max_wait_time = 600, progress_callback = nil)
  logger.info "[assemblyai:poll] Polling for completion: #{transcript_id}"

  start_time = Time.current
  poll_interval = 10 # seconds

  loop do
    result = get_transcription(transcript_id)
    status = result['status']

    case status
    when 'completed'
      logger.info '[assemblyai:poll] Transcription completed successfully'
      return result
    when 'error'
      error_msg = result['error'] || 'Unknown error'
      logger.error "[assemblyai:poll] Transcription failed: #{error_msg}"
      raise "AssemblyAI transcription failed: #{error_msg}"
    when 'queued', 'processing'
      elapsed_time = Time.current - start_time

      if elapsed_time > max_wait_time
        logger.error "[assemblyai:poll] Timeout waiting for transcription completion after #{elapsed_time.to_i} seconds (max: #{max_wait_time}s)"
        raise "AssemblyAI transcription timeout after #{elapsed_time.to_i} seconds. Video may be too long or AssemblyAI service is experiencing delays. Please try again later."
      end

      logger.info "[assemblyai:poll] Status: #{status}, waiting #{poll_interval}s... (elapsed: #{elapsed_time.to_i}s / max: #{max_wait_time}s)"

      # Call progress callback if provided
      progress_callback&.call(status, elapsed_time)

      sleep(poll_interval)
    else
      logger.error "[assemblyai:poll] Unknown status: #{status}"
      raise "AssemblyAI unknown transcription status: #{status}"
    end
  end
end

#submit_transcription(upload_url, options = {}) ⇒ String

Submit a transcription request

Parameters:

  • upload_url (String) —

    The audio URL to transcribe

  • options (Hash) (defaults to: {}) —

    Transcription options (passed through to AssemblyAI API)

Returns:

  • (String) —

    The transcript ID



49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
 
# File 'app/services/assemblyai_client.rb', line 49

def submit_transcription(upload_url, options = {})
  logger.info "[assemblyai:transcribe] Submitting transcription for: #{upload_url}"

  transcript_url = "#{BASE_URL}/transcript"

  # Start with defaults, then merge caller options
  # This allows callers to override or add any AssemblyAI API option
  defaults = {
    language_code: 'en_us',
    punctuate: true,
    format_text: true,
    speaker_labels: true # Enable speaker diarization by default
  }

  # Build request body: defaults -> caller options -> required audio_url
  request_body = defaults.merge(options).merge(audio_url: upload_url)

  # Remove nil values (AssemblyAI doesn't like them)
  request_body.compact!

  logger.debug("[assemblyai:transcribe] Request sent#{' with webhook' if options[:webhook_url].present?}")

  begin
    response = @client.post(transcript_url, request_body.to_json)

    if response.status == 200
      result = JSON.parse(response.body)
      logger.info "[assemblyai:transcribe] Success: #{result['id']}"
      result['id']
    else
      logger.error "[assemblyai:transcribe] Failed with status #{response.status}: #{response.body}"
      raise "AssemblyAI transcription submission failed: #{response.status} - #{response.body}"
    end
  rescue StandardError => e
    logger.error "[assemblyai:transcribe] Exception: #{e.message}"
    raise "AssemblyAI transcription submission failed: #{e.message}"
  end
end

#transcribe_file(file_path, options = {}) ⇒ Object

Complete transcription workflow: upload + transcribe + poll



152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
 
# File 'app/services/assemblyai_client.rb', line 152

def transcribe_file(file_path, options = {})
  logger.info "[assemblyai:transcribe_file] Starting transcription workflow for: #{file_path}"

  # Step 1: Upload file
  upload_url = upload_file(file_path)

  # Step 2: Submit transcription
  transcript_id = submit_transcription(upload_url, options)

  # Step 3: Poll for completion
  result = poll_transcription(transcript_id, options[:max_wait_time])

  # Step 4: Format result and include transcript ID
  formatted_result = format_transcription_result(result)
  formatted_result['id'] = transcript_id

  formatted_result
end

#translate_transcript(transcript_id:, target_languages:, formal: true, match_utterances: true, timeout: 300) ⇒ Hash

Native translation using AssemblyAI Speech Understanding API
Translates transcript with utterance timing preserved in a single API call

Parameters:

  • transcript_id (String) —

    AssemblyAI transcript ID

  • target_languages (Array<String>) —

    Target language codes (e.g., ['es', 'fr', 'pl'])

  • formal (Boolean) (defaults to: true) —

    Use formal language style (default: true)

  • match_utterances (Boolean) (defaults to: true) —

    Preserve original utterance boundaries (default: true)

  • timeout (Integer) (defaults to: 300) —

    Request timeout in seconds (default: 300 for long videos)

Returns:

  • (Hash) —

    Response with translated texts and utterances



440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
 
# File 'app/services/assemblyai_client.rb', line 440

def translate_transcript(transcript_id:, target_languages:, formal: true, match_utterances: true, timeout: 300)
  logger.info "[assemblyai:translate_transcript] Translating transcript #{transcript_id} to #{target_languages.join(', ')}"

  request_body = {
    transcript_id: transcript_id,
    speech_understanding: {
      request: {
        translation: {
          target_languages: target_languages,
          formal: formal,
          match_original_utterance: match_utterances
        }
      }
    }
  }

  begin
    # Use longer timeout for translation of long videos
    response = @client.post("#{LLM_GATEWAY_URL}/understanding", request_body.to_json, nil, nil, timeout)

    if response.status == 200
      result = JSON.parse(response.body)
      logger.info "[assemblyai:translate_transcript] Success - translated to #{target_languages.length} languages"

      # Log utterance count if available
      if result['utterances'].present?
        logger.info "[assemblyai:translate_transcript] Utterances returned: #{result['utterances'].length}"
      end

      result
    else
      error_body = response.body.to_s.truncate(500)
      logger.error "[assemblyai:translate_transcript] Failed with status #{response.status}: #{error_body}"
      raise "Speech Understanding translation failed: #{response.status} - #{error_body}"
    end
  rescue StandardError => e
    logger.error "[assemblyai:translate_transcript] Exception: #{e.message}"
    raise "Speech Understanding translation failed: #{e.message}"
  end
end

#upload_file(file_path) ⇒ Object

Upload a file to AssemblyAI and get upload URL



23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
 
# File 'app/services/assemblyai_client.rb', line 23

def upload_file(file_path)
  logger.info "[assemblyai:upload] Uploading file: #{file_path}"

  upload_url = "#{BASE_URL}/upload"

  begin
    response = @client.post(upload_url, File.read(file_path))

    if response.status == 200
      result = JSON.parse(response.body)
      logger.info "[assemblyai:upload] Success: #{result['upload_url']}"
      result['upload_url']
    else
      logger.error "[assemblyai:upload] Failed with status #{response.status}: #{response.body}"
      raise "AssemblyAI upload failed: #{response.status} - #{response.body}"
    end
  rescue StandardError => e
    logger.error "[assemblyai:upload] Exception: #{e.message}"
    raise "AssemblyAI upload failed: #{e.message}"
  end
end